how to write a research paper using ai

AI can write a passing college paper in 20 minutes

AI can do a lot of things extremely well. One thing that it can do just okay -- which, frankly, is still quite extraordinary -- is write college term papers.

That's the finding from EduRef , a resource for students and educators, which ran an experiment to determine if a deep learning language prediction model known as GPT-3 could get passing marks in an anonymized trial.

Also: What is GPT-3? Everything you need to know about OpenAI's breakthrough AI language program

• iPhone 15: Four things make this a fun upgrade year
• How to use ChatGPT to make charts and tables
• Every product we're expecting at Apple's September event (and what won't be unveiled)
• The best phone deals right now: Save on the latest iPhones and Androids

"We hired a panel of professors to create a writing prompt, gave it to a group of recent grads and undergraduate-level writers, and fed it to GPT-3  and had the panel grade the anonymous submissions and complete a follow up survey for thoughts about the writers," according to an EduRef post. The results were a surprising demonstration of the natural-language prowess of AI .

The specific AI -- GPT-3, for Generative Pre-trained Transformer 3 -- was released in June 2020 by OpenAI , a research business co-founded by Elon Musk. It was developed to create content with a human language structure better than any of its predecessors. Natural language processing has been developing swiftly in the past couple years, enabling computers to generate text that feels, in many cases, contextually appropriate and passably organic. 

However, the hurdles for advanced natural language processing are enormous. According to a 2019 paper by the Allen Institute of Artificial Intelligence, machines fundamentally lack commonsense reasoning -- the ability to understand what they're writing. That finding is based on a critical reevaluation of standard tests to determine commonsense reasoning in machines, such as the  Winograd Schema Challenge .

Which makes the results of the EduRef experiment that much more striking. The writing prompts were given in a variety of subjects, including U.S. History, Research Methods (Covid-19 Vaccine Efficacy), Creative Writing, and Law. GPT-3 managed to score a "C" average across four subjects from professors, failing only one assignment. The AI scored the highest grades in U.S. History and Law writing prompts, earning a B- in both assignments. GPT-3 scored a "C" in a research paper on Covid-19 Vaccine Efficacy, scoring better than one human writer.

Overall, the instructor evaluations suggested that writing produced by GPT-3 was able to mimic human writing in areas of grammar, syntax, and word frequency, although the papers felt somewhat technical. As you might expect, the time it took the AI to complete the assignments was dramatically less than that required by human participants. The average time between assignment and completion for humans was 3 days, while the average time between assignment and completion for GPT-3 was between 3 and 20 minutes. 

"Even without being augmented by human interference, GPT-3's assignments received more or less the same feedback as the human writers," according to EduRef. "While 49.2% of comments on GPT-3's work were related to grammar and syntax, 26.2% were about focus and details. Voice and organization were also mentioned, but only 12.3% and 10.8% of the time, respectively. Similarly, our human writers received comments in nearly identical proportions. Almost 50% of comments on the human papers were related to grammar and syntax, with 25.4% related to focus and details. Just over 13% of comments were about the humans' use of voice, while 10.4% were related to organization."

Aside from potentially troubling implications for educators, what this points to is a dawning inflection point for natural language processing, heretofore a decidedly human characteristic.

Artificial Intelligence

Nearly 40% of workers think generative AI can help with workplace communication

Google debuts Duet AI to tackle new cybersecurity challenges in the cloud

OpenAI finally introduces a business version of ChatGPT

How would you rate Smodin?

Help us improve Smodin by leaving us feedback!

World’s #1 Research Paper Generator

Over 5,000 research papers generated daily

Have an AI Research and write your Paper in just 5 words

See it for yourself: get your free research paper started with just 5 words, how smodin makes research paper writing easy, instantly find sources for any sentence.

Our AI research tool in the research paper editor interface makes it easy to find a source or fact check any piece of text on the web. It will find you the most relevant or related piece of information and the source it came from. You can quickly add that reference to your document references with just a click of a button. We also provide other modes for research such as “find support statistics”, “find supporting arguments”, “find useful information”, and other research methods to make finding the information you need a breeze. Make research paper writing and research easy with our AI research assistant.

Easily Cite References

Our research paper generator makes citing references in MLA and APA styles for web sources and references an easy task. The research paper writer works by first identifying the primary elements in each source, such as the author, title, publication date, and URL, and then organizing them in the correct format required by the chosen citation style. This ensures that the references are accurate, complete, and consistent. The product provides helpful tools to generate citations and bibliographies in the appropriate style, making it easier for you to document your sources and avoid plagiarism. Whether you’re a student or a professional writer, our research paper generator saves you time and effort in the citation process, allowing you to focus on the content of your work.

Free AI Research Paper Generator & Writer - Say Goodbye to Writer's Block!

Are you struggling with writer's block? Even more so when it comes to your research papers. Do you want to write a paper that excels, but can't seem to find the inspiration to do so? Say goodbye to writer's block with Smodin’s Free AI Research Paper Generator & Writer!

Smodin’s AI-powered tool generates high-quality research papers by analyzing millions of papers and using advanced algorithms to create unique content. All you need to do is input your topic, and Smodin’s Research Paper generator will provide you with a well-written paper in no time.

Why Use Smodin Free AI Research Paper Generator & Writer?

Writing a research paper can be a complicated task, even more so when you have limited time and resources. A research paper generator can help you streamline the process, by quickly finding and organizing relevant sources. With Smodin's research paper generator, you can produce high-quality papers in minutes, giving you more time to focus on analysis and writing

Benefits of Smodin’s Free Research Paper Generator

• Save Time: Smodin AI-powered generator saves you time by providing you with a well-written paper that you can edit and submit.
• Quality Content: Smodin uses advanced algorithms to analyze millions of papers to ensure that the content is of the highest quality.
• Easy to Use: Smodin is easy to use, even if you're not familiar with the topic. It is perfect for students, researchers, and professionals who want to create high-quality content.

How to Write a Research Paper?

All you need is an abstract or a title, and Smodin’s AI-powered software will quickly find sources for any topic or subject you need. With Smodin, you can easily produce multiple sections, including the introduction, discussion, and conclusion, saving you valuable time and effort.

Who can write a Research Paper?

Everyone can! Smodin's research paper generator is perfect for students, researchers, and anyone else who needs to produce high-quality research papers quickly and efficiently. Whether you're struggling with writer's block or simply don't have the time to conduct extensive research, Smodin can help you achieve your goals.

Tips for Using Smodin's Research Paper Generator

With our user-friendly interface and advanced AI algorithms, you can trust Smodin's paper writer to deliver accurate and reliable results every time. While Smodin's research paper generator is designed to be easy to use, there are a few tips you can follow to get the most out of Smodinl. First, be sure to input a clear and concise abstract or title to ensure accurate results. Second, review and edit the generated paper to ensure it meets your specific requirements and style. And finally, use the generated paper as a starting point for your research and writing, or to continue generating text.

The Future of Research Paper Writing

As technology continues to advance, the future of research paper writing is likely to become increasingly automated. With tools like Smodin's research paper generator, researchers and students can save time and effort while producing high-quality work. Whether you're looking to streamline your research process or simply need a starting point for your next paper, Smodin's paper generator is a valuable resource for anyone interested in academic writing.

So why wait? Try Smodin's free AI research paper generator and paper writer today and experience the power of cutting-edge technology for yourself. With Smodin, you can produce high-quality research papers in minutes, saving you time and effort while ensuring your work is of the highest caliber.

© 2023 Smodin LLC

How to Write a Research Paper (+Free AI Research Paper Writer)

Over the years, I've managed to vastly improve how I write research papers.

The three major game-changers for me, in terms of quality of the finished piece, have been:

• Following the research paper checklist (see below)
• Developing the thesis before starting to write
• And, more recently, using AI to improve my research paper draft

Let's break down each of these elements and produce the kind of research papers that get cited in magazines.

FREE AI research paper writer > FREE AI research paper writer >

What is a research paper, and how is it written differently?

Research papers are longer and more in-depth than essays. They require extensive research and evidence-based arguments. Research papers also typically have a more formal structure and require citations and references.When academics want to find a balanced and comprehensive view on a given topic, they usually seek a research paper.

Like most writing assignments, a research paper can be broken down into simple steps. Research papers follow the same basic writing process as explanatory or persuasive essays — but instead of making an argument or drawing greater meaning from the topic, the research paper is primarily concerned with concrete facts that may be analyzed, examined, or interpreted to better understand the paper’s central topic.

This is good news if you enjoy research: you’ll be doing a lot of it. The ultimate quality of your paper depends on you conducting thorough, complete research — and relying on reputable sources.

1. Make a checklist based on the assignment description, and fill it out with AI

Your professor has likely specified some criteria for your research paper:

• Length (in pages or words)
• Type of topic (the War of 1812, ancient Greece, agriculture, etc.)
• Elements that must be included, such as analysis, discussion, and comparison.
• Types of sources you must draw from (academic papers, encyclopedias, etc.)
• Source attribution style
• Formatting style

Go through the assignment description and create a checklist of those criteria. You can use this checklist throughout the research and writing process as well:

AI can really help you get some traction with your research paper in the preperation stage. This includes two main steps:

• Brainstorming paper topic idea
• Outlining based on your topic, basing the prompt on the assignment

2. Choose a topic you’re curious about, or use AI to help you with that

A sure way to write a boring research paper is to pick a topic you have no interest in, like summer temperatures in the desert or the life cycle of a flea. (Though someone’s probably interested in those things.)

Instead, follow your curiosity.

If your paper is for a writing class, you may have a lot of freedom to choose what you write about, so tap into your interests. Are you intrigued by the history of roller skating or the invention of the soccer cleat? Or how teen social dynamics have changed with evolving technology (think: home phones → online instant messaging → flip phones → smartphones)?

If you’re writing for a class in a subject like history, art, or science, you’ll probably have more restrictions on what you can write about — like a time period or type of art or science — but you can still use your curiosity to pick an interesting topic.

If you’re having a tough time, try brainstorming a list of things you’ve wondered about. Ask “ what’s up with… ” and see what comes to mind.

For example:

What’s up with traffic circles and why are they supposedly better for traffic patterns than a light or four-way stop?

What’s up with country music sounding more and more like hip-hop?

What’s up with people who have gluten allergies being able to eat bread in Europe but not the US?

Once you have a list, choose the topic you find most interesting (and appropriate for the assignment).

If your mind draws a blank, you can utilize AI to help you choose a topic. Let's say your course is about mid century art. You can go to a tool like Wotdtune and ask it to give you ideas for creative mid century art essays. See example below.

3. Develop your thesis (and guide your research) by asking a research question

Even though a research paper may not necessarily take a side on a topic, it still needs a thesis, aka a central idea or focus that drives the piece from beginning to end. 

We wrote a whole guide on writing thesis statements , so here we’ll just give you this tip:

Use a research question to develop your thesis

A research question is a variation on the “What’s up with…” questions from the last tip — but it will zoom in more specifically on the aspect of your topic that you’re investigating.

Why were the Irish so dependent on potatoes?

Did any women in ancient Greece enjoy relative freedom and autonomy?

You may already know the answer to these questions, or you may not. Either way, they give you a place to start in your research. Once you have your question, set out to:

• Find the initial answer.
• Gather more context (the who, what, when, where, why, how) around that answer.
• Revise your research question and turn it into your thesis.

This process helps tighten your focus from a broad topic that could fill books to a specific angle that can be meaningfully explored in the few pages of your paper.

Instead of the potato famine , write about why England was to blame for the potato famine’s devastating effects on the Irish.

Instead of ancient Greece or women in ancient Greece , write about how Spartan women’s lives differed from the lives of women in Athens.

4. Skim sources and use AI to perform research for your paper

Your research question can help you quickly determine whether information is relevant to your paper. As you gather initial sources, skim them — and then use your research question to decide whether to keep or discard the source. 

Does the source cover information relevant to my research question?

Yes: Keep to read later.

No: Discard and move on to the next source.

This approach will save you precious research time. You won’t waste limited hours reading sources that don’t have a single helpful fact.

If skimming is hard for you (as a deep reader, I get it), Wordtune Read can help. Paste the link to your online source, upload a scanned PDF, or copy the text, and the tool will scan and summarize for you. You can always come back later and closely read the most useful sources.

5. Make note of the most interesting facts you find.

Along with taking detailed notes of your research (complete with all the source info you need to make proper citations), highlight the most interesting facts you come across. You could stick these in a section together or mark them in a way that makes them stand out.

Why should you do this?

Because later on, one of these fascinating factoids could have a direct connection to your thesis — and make a great hook for the start of your paper. Instead of digging through all of your notes to try to remember what that interesting tidbit was, you’ll be able to find it easily.

6. Organize your research.

There are plenty of ways to organize your notes, but I suggest breaking them up into subtopics and categories.

• Subtopic: A topic related to your main topic or thesis that needs to be explained and understood by readers in order to understand your main topic or thesis. For example: Land ownership in Ireland under British rule.
• Category: An overarching concept that several subtopics fall under. For example: British restrictions on the Irish.

To start, I would focus on the subtopics and then group them into categories.

As you organize, use the formatting tools in your word processor to tag headings and subheadings. For example, all categories would be an H2 (Heading 2), while all subtopics would be an H3 (Heading 3). 

Tagging your categories and subtopics this way will help you develop your outline. Just organize your categories and subtopics in a logical order, and you’ll have a skeleton of an outline ready to go.

7. Write with your research document open

No one can remember everything they found while researching — you’ll need to reference your research document throughout the writing process. No question there.

But you can make this easier (and keep your writing process efficient) by:

Keeping your research document open and in clear view.

I like to put my draft document and my research document side by side on my screen, so I can see them both at the same time. 

Another approach would be to paste the information you need directly into your draft document — in the order you’ll need it. (Your outline will help you know what you need.)

8. Steal the TK trick from journalists

In the middle of drafting your paper, you find that you’re missing a fact. 

You neglected to write down how many Irish people starved due to the potato famine.

You don’t know what age Spartan women were able to own property.

Instead of derailing your writing and searching for that information, write the sentence you want to write and stick a “TK” where the missing fact should go.

“TK” stands for “to come” (don't ask us why) and is a placeholder used by journalists to mark missing information they’ll fill in later. Using TK allows you to keep writing without getting off track every time you discover your research didn’t cover everything.

A whopping TK Irish people starved, thanks to the combination of famine and British oppression.

At age TK , Spartan girls became women who were able to own property, a right that their sisters in Athens did not enjoy.

9. Revise, explain, paraphrase with AI as your research/writing assistant

Using the right researching tools can get you a lot way.

If you’re ever at a loss for words — writing clunky, clumsy sentences, struggling to explain a concept, or having a hard time paraphrasing a source — Wordtune can serve as your AI sidekick. 

Simply highlight the sentence in question and browse Wordtune’s suggestion for a better wording.

You can also use Wordtune Spices to come up with examples and counter arguments for whatever you're writing about or even find stats and facts, complete with source citations

Wordtune doesn’t do all of the writing for you, but it can help you sharpen your ideas on the sentence level, so you can hand in a research paper with good writing that’s still very much your own.

P.S. This article was co-written with Wordtune . Wordtune didn’t write the whole piece. Instead, it contributed ideas, suggested rephrasing alternatives, maintained consistency in tone, and of course - made the process much more fun for the writer.

Join our newsletter and get better at writing

Thanks you’re in..

6 Ways AI Writing Tools Can Save You Time (and 4 Ways That they Can’t)

Discover how AI writing tools can revolutionize your writing process and save you valuable time. Explore six powerful ways to streamline your content creation with artificial intelligence.

8 Websites Every Copywriter Must Know in 2023

Looking to launch a career as a successful freelance copywriter or develop your writing skills further? Here are the 8 best websites for copywriters in 2023.

How to Paraphrase Without Plagiarizing (Using AI)

Learn the art of AI-powered paraphrasing to create original content and evade plagiarism pitfalls. Unlock the secrets of effective rephrasing with insightful guide on harnessing AI for plagiarism-free paraphrasing in your writing endeavors.

Artificial Intelligence in Cardiothoracic Imaging pp 567–574 Cite as

How to Write and Review an Artificial Intelligence Paper

• Thomas Weikert 5 &
• Tim Leiner 6  
• First Online: 22 April 2022

1257 Accesses

Part of the Contemporary Medical Imaging book series (CMI)

The purpose of this chapter is to provide medical imaging professionals with the tools to write a research article in the field of artificial intelligence. At the same time, this can help readers to assess the quality of a publication. To this end, the chapter discusses 12 key considerations in detail, ranging from defining a research objective to public sharing of software code. Furthermore, a checklist of 25 items based on the standard structure of a research articles is derived from these considerations to provide writers and readers with an easily applicable tool.

• Cardiovascular
• Artificial intelligence
• Machine learning
• Quality standards

This is a preview of subscription content, access via your institution .

Buying options

• Available as PDF
• Read on any device
• Instant download
• Own it forever
• Available as EPUB and PDF
• Compact, lightweight edition
• Dispatched in 3 to 5 business days
• Free shipping worldwide - see info
• Durable hardcover edition

Tax calculation will be finalised at checkout

Purchases are for personal use only

Abadi M, Agarwal A, Barham P, Brevdo E, Chen Z, Citro C, Corrado GS, Davis A, Dean J, Devin M, Ghemawat S, Goodfellow I, Harp A, Irving G, Isard M, Jia Y, Jozefowicz R, Kaiser L, Kudlur M, Zheng X. TensorFlow: large-scale machine learning on heterogeneous distributed systems. 2016.

Google Scholar  

Balki I, Amirabadi A, Levman J, Martel AL, Emersic Z, Meden B, Garcia-Pedrero A, Ramirez SC, Kong D, Moody AR, Tyrrell PN. Sample-size determination methodologies for machine learning in medical imaging research: a systematic review. Canadian Associat Radiolog J. 2019;70(4):344–53). Canadian Medical Association. https://doi.org/10.1016/j.carj.2019.06.002 .

CrossRef   Google Scholar  

Bluemke DA, Moy L, Bredella MA, Ertl-Wagner BB, Fowler KJ, Goh VJ, Halpern EF, Hess CP, Schiebler ML, Weiss CR. Assessing radiology research on artificial intelligence: a brief guide for authors, reviewers, and readers—from the radiology editorial board. Radiology. 2020;294(3):487–9. https://doi.org/10.1148/radiol.2019192515 .

CrossRef   PubMed   Google Scholar  

Bossuyt PM, Reitsma JB, Bruns DE, Gatsonis CA, Glasziou PP, Irwig L, Lijmer JG, Moher D, Rennie D, de Vet HCW, Kressel HY, Rifai N, Golub RM, Altman DG, Hooft L, Korevaar DA, Cohen JF, STARD Group. STARD 2015: an updated list of essential items for reporting diagnostic accuracy studies. Radiology. 2015;277(3):826–32. https://doi.org/10.1148/radiol.2015151516 .

Chan A-W, Tetzlaff JM, Altman DG, Laupacis A, Gøtzsche PC, Krleža-Jerić K, Hróbjartsson A, Mann H, Dickersin K, Berlin JA, Doré CJ, Parulekar WR, Summerskill WSM, Groves T, Schulz KF, Sox HC, Rockhold FW, Rennie D, Moher D. SPIRIT 2013 statement: defining standard protocol items for clinical trials. Ann Intern Med. 2013;158(3):200–7. https://doi.org/10.7326/0003-4819-158-3-201302050-00583 .

CrossRef   PubMed   PubMed Central   Google Scholar  

Collins GS, Reitsma JB, Altman DG, Moons KGM. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. BMJ. 2015;350(jan07 4):g7594. https://doi.org/10.1136/bmj.g7594 .

FDA. Proposed regulatory framework for modifications to Artificial Intelligence/Machine Learning (AI/ML)-Based Software as a Medical Device (SaMD) - Discussion Paper and Request for Feedback. 2019.

Mongan J, Moy L, Kahn JEJ. Checklist for artificial intelligence in medical imaging (CLAIM): a guide for authors and reviewers. Radiology: Artificial Intelligence. 2020:2. https://doi.org/10.1148/ryai.2020200029 .

Paszke A, Gross S, Massa F, Lerer A, Bradbury J, Chanan G, Killeen T, Lin Z, Gimelshein N, Antiga L, Desmaison A, Köpf A, Yang E, DeVito Z, Raison M, Tejani A, Chilamkurthy S, Steiner B, Fang L, et al. PyTorch: an imperative style. High-Performance Deep Learning Library. 2019; http://arxiv.org/abs/1912.01703 .

Pedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B, Grisel O, Blondel M, Prettenhofer P, Weiss R, Dubourg V, Vanderplas J, Passos A, Cournapeau D, Brucher M, Perrot M, Duchesnay É. Scikit-learn: machine learning in python. J Mach Learn Res. 2011;12(Oct):2825–30.

Download references

Author information

Authors and affiliations.

Department of Radiology, University Hospital Basel, University of Basel, Basel, Switzerland

Thomas Weikert

Department of Radiology, Mayo Clinic, Rochester, MN, USA

You can also search for this author in PubMed   Google Scholar

Corresponding author

Correspondence to Thomas Weikert .

Editor information

Editors and affiliations.

Division of Cardiothoracic Imaging, Department of Radiology and Imaging Sciences, Emory University, Atlanta, GA, USA

MD Carlo N. De Cecco

Ph.D. Marly van Assen

MD Tim Leiner

Rights and permissions

Reprints and Permissions

Copyright information

© 2022 The Author(s), under exclusive license to Springer Nature Switzerland AG

About this chapter

Cite this chapter.

Weikert, T., Leiner, T. (2022). How to Write and Review an Artificial Intelligence Paper. In: De Cecco, C.N., van Assen, M., Leiner, T. (eds) Artificial Intelligence in Cardiothoracic Imaging. Contemporary Medical Imaging. Humana, Cham. https://doi.org/10.1007/978-3-030-92087-6_53

Download citation

DOI : https://doi.org/10.1007/978-3-030-92087-6_53

Published : 22 April 2022

Publisher Name : Humana, Cham

Print ISBN : 978-3-030-92086-9

Online ISBN : 978-3-030-92087-6

eBook Packages : Medicine Medicine (R0)

Share this chapter

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Find a journal
• Publish with us

Have a language expert improve your writing

Run a free plagiarism check in 10 minutes, generate accurate citations for free.

• Knowledge Base
• Research paper

How to Write a Research Paper | A Beginner's Guide

A research paper is a piece of academic writing that provides analysis, interpretation, and argument based on in-depth independent research.

Research papers are similar to academic essays , but they are usually longer and more detailed assignments, designed to assess not only your writing skills but also your skills in scholarly research. Writing a research paper requires you to demonstrate a strong knowledge of your topic, engage with a variety of sources, and make an original contribution to the debate.

This step-by-step guide takes you through the entire writing process, from understanding your assignment to proofreading your final draft.

Table of contents

Understand the assignment, choose a research paper topic, conduct preliminary research, develop a thesis statement, create a research paper outline, write a first draft of the research paper, write the introduction, write a compelling body of text, write the conclusion, the second draft, the revision process, research paper checklist, free lecture slides.

Completing a research paper successfully means accomplishing the specific tasks set out for you. Before you start, make sure you thoroughly understanding the assignment task sheet:

• Read it carefully, looking for anything confusing you might need to clarify with your professor.
• Identify the assignment goal, deadline, length specifications, formatting, and submission method.
• Make a bulleted list of the key points, then go back and cross completed items off as you’re writing.

Carefully consider your timeframe and word limit: be realistic, and plan enough time to research, write, and edit.

Here's why students love Scribbr's proofreading services

Discover proofreading & editing

There are many ways to generate an idea for a research paper, from brainstorming with pen and paper to talking it through with a fellow student or professor.

You can try free writing, which involves taking a broad topic and writing continuously for two or three minutes to identify absolutely anything relevant that could be interesting.

You can also gain inspiration from other research. The discussion or recommendations sections of research papers often include ideas for other specific topics that require further examination.

Once you have a broad subject area, narrow it down to choose a topic that interests you, m eets the criteria of your assignment, and i s possible to research. Aim for ideas that are both original and specific:

• A paper following the chronology of World War II would not be original or specific enough.
• A paper on the experience of Danish citizens living close to the German border during World War II would be specific and could be original enough.

Note any discussions that seem important to the topic, and try to find an issue that you can focus your paper around. Use a variety of sources , including journals, books, and reliable websites, to ensure you do not miss anything glaring.

Do not only verify the ideas you have in mind, but look for sources that contradict your point of view.

• Is there anything people seem to overlook in the sources you research?
• Are there any heated debates you can address?
• Do you have a unique take on your topic?
• Have there been some recent developments that build on the extant research?

In this stage, you might find it helpful to formulate some research questions to help guide you. To write research questions, try to finish the following sentence: “I want to know how/what/why…”

A thesis statement is a statement of your central argument — it establishes the purpose and position of your paper. If you started with a research question, the thesis statement should answer it. It should also show what evidence and reasoning you’ll use to support that answer.

The thesis statement should be concise, contentious, and coherent. That means it should briefly summarize your argument in a sentence or two, make a claim that requires further evidence or analysis, and make a coherent point that relates to every part of the paper.

You will probably revise and refine the thesis statement as you do more research, but it can serve as a guide throughout the writing process. Every paragraph should aim to support and develop this central claim.

A research paper outline is essentially a list of the key topics, arguments, and evidence you want to include, divided into sections with headings so that you know roughly what the paper will look like before you start writing.

A structure outline can help make the writing process much more efficient, so it’s worth dedicating some time to create one.

Your first draft won’t be perfect — you can polish later on. Your priorities at this stage are as follows:

• Maintaining forward momentum — write now, perfect later.
• Paying attention to clear organization and logical ordering of paragraphs and sentences, which will help when you come to the second draft.
• Expressing your ideas as clearly as possible, so you know what you were trying to say when you come back to the text.

You do not need to start by writing the introduction. Begin where it feels most natural for you — some prefer to finish the most difficult sections first, while others choose to start with the easiest part. If you created an outline, use it as a map while you work.

Do not delete large sections of text. If you begin to dislike something you have written or find it doesn’t quite fit, move it to a different document, but don’t lose it completely — you never know if it might come in useful later.

Paragraph structure

Paragraphs are the basic building blocks of research papers. Each one should focus on a single claim or idea that helps to establish the overall argument or purpose of the paper.

Example paragraph

George Orwell’s 1946 essay “Politics and the English Language” has had an enduring impact on thought about the relationship between politics and language. This impact is particularly obvious in light of the various critical review articles that have recently referenced the essay. For example, consider Mark Falcoff’s 2009 article in The National Review Online, “The Perversion of Language; or, Orwell Revisited,” in which he analyzes several common words (“activist,” “civil-rights leader,” “diversity,” and more). Falcoff’s close analysis of the ambiguity built into political language intentionally mirrors Orwell’s own point-by-point analysis of the political language of his day. Even 63 years after its publication, Orwell’s essay is emulated by contemporary thinkers.

Citing sources

It’s also important to keep track of citations at this stage to avoid accidental plagiarism . Each time you use a source, make sure to take note of where the information came from.

You can use our free citation generators to automatically create citations and save your reference list as you go.

APA Citation Generator MLA Citation Generator

The research paper introduction should address three questions: What, why, and how? After finishing the introduction, the reader should know what the paper is about, why it is worth reading, and how you’ll build your arguments.

What? Be specific about the topic of the paper, introduce the background, and define key terms or concepts.

Why? This is the most important, but also the most difficult, part of the introduction. Try to provide brief answers to the following questions: What new material or insight are you offering? What important issues does your essay help define or answer?

How? To let the reader know what to expect from the rest of the paper, the introduction should include a “map” of what will be discussed, briefly presenting the key elements of the paper in chronological order.

The major struggle faced by most writers is how to organize the information presented in the paper, which is one reason an outline is so useful. However, remember that the outline is only a guide and, when writing, you can be flexible with the order in which the information and arguments are presented.

One way to stay on track is to use your thesis statement and topic sentences . Check:

• topic sentences against the thesis statement;
• topic sentences against each other, for similarities and logical ordering;
• and each sentence against the topic sentence of that paragraph.

Be aware of paragraphs that seem to cover the same things. If two paragraphs discuss something similar, they must approach that topic in different ways. Aim to create smooth transitions between sentences, paragraphs, and sections.

The research paper conclusion is designed to help your reader out of the paper’s argument, giving them a sense of finality.

Trace the course of the paper, emphasizing how it all comes together to prove your thesis statement. Give the paper a sense of finality by making sure the reader understands how you’ve settled the issues raised in the introduction.

You might also discuss the more general consequences of the argument, outline what the paper offers to future students of the topic, and suggest any questions the paper’s argument raises but cannot or does not try to answer.

You should not :

• Offer new arguments or essential information
• Take up any more space than necessary
• Begin with stock phrases that signal you are ending the paper (e.g. “In conclusion”)

There are four main considerations when it comes to the second draft.

• Check how your vision of the paper lines up with the first draft and, more importantly, that your paper still answers the assignment.
• Identify any assumptions that might require (more substantial) justification, keeping your reader’s perspective foremost in mind. Remove these points if you cannot substantiate them further.
• Be open to rearranging your ideas. Check whether any sections feel out of place and whether your ideas could be better organized.
• If you find that old ideas do not fit as well as you anticipated, you should cut them out or condense them. You might also find that new and well-suited ideas occurred to you during the writing of the first draft — now is the time to make them part of the paper.

The goal during the revision and proofreading process is to ensure you have completed all the necessary tasks and that the paper is as well-articulated as possible.

Global concerns

• Confirm that your paper completes every task specified in your assignment sheet.
• Check for logical organization and flow of paragraphs.
• Check paragraphs against the introduction and thesis statement.

Fine-grained details

Check the content of each paragraph, making sure that:

• each sentence helps support the topic sentence.
• no unnecessary or irrelevant information is present.
• all technical terms your audience might not know are identified.

Next, think about sentence structure , grammatical errors, and formatting . Check that you have correctly used transition words and phrases to show the connections between your ideas. Look for typos, cut unnecessary words, and check for consistency in aspects such as heading formatting and spellings .

Finally, you need to make sure your paper is correctly formatted according to the rules of the citation style you are using. For example, you might need to include an MLA heading  or create an APA title page .

Scribbr’s professional editors can help with the revision process with our award-winning proofreading services.

Discover our paper editing service

Checklist: Research paper

I have followed all instructions in the assignment sheet.

My introduction presents my topic in an engaging way and provides necessary background information.

My introduction presents a clear, focused research problem and/or thesis statement .

My paper is logically organized using paragraphs and (if relevant) section headings .

Each paragraph is clearly focused on one central idea, expressed in a clear topic sentence .

Each paragraph is relevant to my research problem or thesis statement.

I have used appropriate transitions  to clarify the connections between sections, paragraphs, and sentences.

My conclusion provides a concise answer to the research question or emphasizes how the thesis has been supported.

My conclusion shows how my research has contributed to knowledge or understanding of my topic.

My conclusion does not present any new points or information essential to my argument.

I have provided an in-text citation every time I refer to ideas or information from a source.

I have included a reference list at the end of my paper, consistently formatted according to a specific citation style .

I have thoroughly revised my paper and addressed any feedback from my professor or supervisor.

I have followed all formatting guidelines (page numbers, headers, spacing, etc.).

You've written a great paper. Make sure it's perfect with the help of a Scribbr editor!

Open Google Slides Download PowerPoint

Is this article helpful?

Other students also liked.

• Writing a Research Paper Introduction | Step-by-Step Guide
• Writing a Research Paper Conclusion | Step-by-Step Guide
• Research Paper Format | APA, MLA, & Chicago Templates

More interesting articles

• Academic Paragraph Structure | Step-by-Step Guide & Examples
• Checklist: Writing a Great Research Paper
• How to Create a Structured Research Paper Outline | Example
• How to Write a Discussion Section | Tips & Examples
• How to Write Recommendations in Research | Examples & Tips
• How to Write Topic Sentences | 4 Steps, Examples & Purpose
• Research Paper Appendix | Example & Templates
• Research Paper Damage Control | Managing a Broken Argument
• What Is a Theoretical Framework? | Guide to Organizing

What is your plagiarism score?

The best AI writing generators in 2023

These 7 ai writing tools will take your content to the next level..

Of course, all AI writing software needs human supervision to deliver the best results. Left to its own devices, it tends to produce fairly generic and frequently incorrect content, even if it can pass for something a human wrote. Now that AI tools are increasingly popular, people also seem more aware of what bland AI-produced content reads like and are likely to spot it—or at least be suspicious of content that feels like it lacks something.

I've been covering this kind of generative AI technology for almost a decade. Since AI is supposedly trying to take my job, I'm somewhat professionally interested in the whole situation. Still, I think I'm pretty safe for now. These AI writing tools are getting incredibly impressive, but you have to work with them, rather than just letting them spit out whatever they want.

So, if you're looking for an AI content generator that will help you write compelling copy, publish blog posts a lot quicker, and otherwise take some of the slow-paced typing out of writing, you've come to the right place. Let's dig in. 

The best AI writing software

Jasper for businesses

Copy.ai for copywriting

Anyword for assisting you with writing

Sudowrite for fiction

Writer for a non-GPT option

Writesonic for GPT-4 content

Rytr for an affordable AI writer

How do AI writing tools work?

Search Google for AI writing software, and you'll find dozens of different options, all with suspiciously similar features. There's a big reason for this: 95% of these AI writing tools use the same large language models (LLMs) as the back end.

Some of the bigger apps are also integrating their own fine-tuning or using other LLMs like Claude. But m ost are really just wrappers connected to OpenAI's GPT-3 and GPT-4 APIs, with a few extra features built on top —even if they try to hide it in their own marketing materials . If you wanted to, you could even create your own version of an AI writing assistant without code using Zapier's OpenAI integrations —that's how much these apps rely on GPT.

See how one writer created an AI writing coach with GPT and other ways you can use OpenAI with Zapier .

Now this isn't to say that none of these AI-powered writing apps are worth using. They all offer a much nicer workflow than ChatGPT or OpenAI's playground , both of which allow you to generate text with GPT as well. And the better apps allow you to set a "voice" or guidelines that apply to all the text you generate. But the difference between these apps isn't really in the quality of their output. With a few exceptions, you'll get very similar results from the same prompt no matter which app you use—even if they use different LLMs. Where the apps on this list stand out is in how easy they make it to integrate AI text generation into an actual workflow.

As for the underlying LLM models themselves, they work by taking a prompt from you, and then predicting what words will best follow on from your request, based on the data they were trained on. That training data includes books, articles, and other documents across all different topics, styles, and genres—and an unbelievable amount of content scraped from the open internet . Basically, LLMs were allowed to crunch through the sum total of human knowledge to form a deep learning neural network—a complex, many-layered, weighted algorithm modeled after the human brain. Yes, that's the kind of thing you have to do to create a computer program that generates bad poems . 

If you want to dive more into the specifics, check out the Zapier articles on natural language processing and how ChatGPT works . But suffice it to say: GPT and other large language models are incredibly powerful already—and because of that, these AI writing tools have a lot of potential. 

What makes the best AI text generator?

How we evaluate and test apps.

All of our best apps roundups are written by humans who've spent much of their careers using, testing, and writing about software. We spend dozens of hours researching and testing apps, using each app as it's intended to be used and evaluating it against the criteria we set for the category. We're never paid for placement in our articles from any app or for links to any site—we value the trust readers put in us to offer authentic evaluations of the categories and apps we review. For more details on our process, read the full rundown of how we select apps to feature on the Zapier blog .

We know that most AI text generators rely on the various versions of GPT, and even those that don't are using very similar models, so most apps aren't going to stand out because of some dramatic difference in the quality of their output. Creating effective, human-like text is now table stakes. It was required for inclusion on this list—but not sufficient on its own.

As I was testing these apps, here's what else I was looking for:

Tools powered by GPT or a similar large language model with well-documented efficacy. In practice, this means that most but not all of the AI writing tools on this list use GPT to a greater or lesser degree. Many apps are starting to hide what models they use and claim to have a lot of secret sauce built on top (because there's a marketing advantage in being different and more powerful), but the reality is that nine times out of ten, it's the GPT API that's doing the heavy lifting.

An interface that gives you a lot of control over the text output. The more options you have to influence the tone, style, language, content, and everything else, the better. I didn't want tools where you just entered a headline and let the AI do the rest; these are all tools that you collaborate with, so you can write great copy quickly.  The best AI writing tools also let you set a default brand voice that's always on.

Ease of use. You shouldn't have to fight to get the AI to do what you want. With AI writing software like this, there will always be some redoing and reshaping to get the exact output you want, but working with the AI shouldn't feel like wrangling a loose horse. Similarly, great help docs and good onboarding were both a major plus. 

Affordability. ChatGPT is currently free, and all these tools are built on top of an API that costs pennies . There was no hard and fast price limit, but the more expensive tools had to justify the extra expense with better features and a nicer app. After all, almost every app will produce pretty similar outputs regardless of what it costs. 

Apps that weren't designed to make spam content. Previous text-generating tools could " spin " content by changing words to synonyms so that unscrupulous website owners could rip off copyrighted material and generally create lots of low-quality, low-value content. None of that on this list.

Even with these criteria, I had more than 40 different AI writing tools to test. Remember: it's relatively easy for a skilled developer to build a wrapper around the GPT API, so I had to dig deep into each one to find out if it was any good or just had a flashy marketing site.

I tested each app by getting it to write a number of different short- and long-form bits of copy, but as expected, there were very few meaningful quality differences. Instead, it was the overall user experience, depth of features, and affordability that determined whether an app made this list. 

The best AI writing generators at a glance

Best ai writing generator for businesses, jasper (web).

Jasper pros:

One of the most mature and feature-filled options on the list 

Integrates with Grammarly, Surfer, and its own AI art generator

Jasper cons:

Expensive given that all the apps use similar language models 

Jasper (formerly Jarvis) is one of the most feature-filled and powerful AI content generators. It was among the first wave of apps built on top of GPT, and its relative longevity means that it feels like a more mature tool than most of the other apps I tested. It 's continued to grow and develop in the months since I first compiled this list.

If you have a business and budget isn't your primary concern, Jasper should be one of the first apps you try. It's pivoted to mostly focus on marketing campaigns rather than just generating generic AI content. That's not a bad thing, but it means that plans now start at $49/month for individual creators and $125/month for teams.

Jasper has also moved away from just being a GPT app. It claims to combine "several large language models" including GPT-4, Claude 2, and PaLM 2, so that "you get the highest quality outputs and superior uptime." While I can't say that I noticed a massive difference between Jasper's output and any other app's, it does give you a few solid controls so that your content matches your brand. 

You can create a brand Voice and Tone by uploading some appropriate sample text. Based on a few examples of my writing, Jasper created a style that "emphasizes a casual, conversational tone with humor, personal anecdotes, listicles, informal language, expertise in various subjects, and a call to action for an engaging and approachable brand voice." I don't think that's a bad summary of the content I fed in, and its output for a few test blog posts like "The Enduring Popularity of Top Gun" felt closer to my writing than when I asked it to use a generic casual tone of voice. Similarly, there's a Knowledge Base where you can add facts about your business and products so Jasper gets important details right. 

While other apps also offer similar features, Jasper's seemed to work better and are fully integrated with the rest of the app. For example, you can create entire marketing campaigns using your custom brand voice. Put a bit of work into fine-tuning it and uploading the right assets to your knowledge base, and I suspect that Jasper really could create some solid first drafts of marketing materials like blog outlines, social media campaign ads, and the like.

Otherwise, Jasper rounds things out with some nice integrations. It has a built-in ChatGPT competitor and AI art generator (though, again, lots of other apps have both), plays nice with the SEO app Surfer , and there's a browser extension to bring Jasper everywhere.

You can also connect Jasper to thousands of other apps using Zapier . Learn more about how to automate Jasper , or try one of the pre-built workflows below.

Create product descriptions in Jasper from new or updated Airtable records

Create Jasper blog posts from new changes to specific column values in monday.com and save the text in Google Docs documents

Run Jasper commands and send Slack channel messages with new pushed messages in Slack

Jasper pricing: Creator plan from $49/month with one brand voice and 50 knowledge assets. Teams plan starts at $125/month for three seats, three brand voices, and 150 knowledge assets.

Best AI writing app for AI copywriting

Copy.ai (web).

Copy.ai pros:

Has an affordable unlimited plan for high-volume users 

Workflow actively solicits your input, which can lead to higher quality content 

Copy.ai cons:

Expensive if you don't produce a lot of content

Pretty much anything Jasper can do, Copy.ai can do too. It has brand voices, an infobase, a chatbot, and team features (though there isn't a browser extension). Consider it the Burger King to Jasper's McDonalds.

And like the Home of the Whopper, Copy.ai appeals to slightly different tastes. While I could argue that Copy.ai has a nicer layout, the reality is it's geared toward a slightly different workflow. While Jasper lets you and the AI loose, Copy.ai slows things down a touch and encourages you to work with its chatbot or use a template that asks some deliberate, probing questions. For creating website copy, social media captions , product descriptions, and similarly specific things, it makes more sense. But for content marketing blog posts and other long-form content, it might annoy you.

The other big difference is the pricing. While both offer plans for $49/month, Copy.ai includes five user seats and unlimited brand voices. For a small team working with multiple brands, it can be a lot cheaper. Also, if you're looking for a free AI writing generator, Copy.ai also offers a free plan that includes 2,000 words per month.

Overall, there are more similarities than differences between Jasper and Copy.ai , and both can create almost all the same kinds of text. Even when it came to analyzing my voice, they both came to pretty similar conclusions. Copy.ai decided that, to mimic me, it had to "focus on creating content that is both educational and entertaining, using a conversational tone that makes readers feel like they're having a chat with a knowledgeable friend" and "not to be afraid to inject some humor or personal anecdotes." If you're in doubt, try them both out and then decide.

Copy.ai also integrates with Zapier , so you can do things like automatically sending content to your CMS or enriching leads straight from your CRM. Here are a couple examples to get you started.

Add new blog posts created with Copy.ai to Webflow

Enrich new Salesforce leads with Copy.ai

Copy.ai pricing: Free for 2,000 words per month; from $49/month for the Pro plan with 5 users and unlimited brand voices.

Best AI writing assistant

Anyword (web).

Anyword pros:

Makes it very easy for you to include specific details, SEO keywords, and other important information 

Engagement scores and other metrics are surprisingly accurate

Anyword cons:

Can be slower to use

Pretty expensive for a more limited set of features than some of the other apps on this list

While you can direct the AI to include certain details and mention specific facts for every app on this list, none make it as easy as Anyword. More than any of the others, the AI here feels like an eager and moderately competent underling that requires a bit of micromanaging ( and can also try to mimic your writing style and brand voice) , rather than a beast that you have to tame with arcane prompts. 

Take one of its main content-generating tools: the Blog Wizard. Like with Copy.ai, the setup process requires you to describe the blog post you want the AI to create and add any SEO keywords you want to target. Anyword then generates a range of titles for you to choose from, along with a predicted engagement score. 

Once you've chosen a title—or written your own—it generates a suggested outline. Approve it, and you get the option for it to create an entire ~2,000-word blog post (boo!) or a blank document where you can prompt it with additional instructions for each section of the outline, telling it things like what facts to mention, what style to take, and what details to cover. There's also a chatbot-like research sidebar that you can ask questions of and solicit input from. While certainly a slower process than most apps, it gives you a serious amount of control over the content you're creating.

Anyword is definitely aimed at marketers, and its other tools—like the Data-Driven Editor and the Website Targeted Message—all allow you to target your content toward specific audiences and give things engagement scores. While I certainly can't confirm the validity of any of these scores, they at least pass the sniff test. I generally thought the AI-generated content that Anyword scored higher was better—and even when I disagreed, I still liked one of the top options. 

Anyword pricing: Starter plan from $49/month for 1 user and 1 brand voice.

Best AI writing tool for writing fiction

Sudowrite (web).

Sudowrite pros:

The only AI tool on the list explicitly aimed at writing fiction 

Super fun to use if you've ever wanted to play around with fiction 

Sudowrite cons:

It's still an AI text generator, so it can produce nonsensical metaphors, clichéd plots, incoherent action, and has a short memory for details 

Very controversial in fiction writing circles

When I saw Sudowrite's marketing copy, I didn't think for a second it would make it onto this list. Then I tried it and…I kind of love it. Sudowrite is a totally different tool than all the others on this list because it's aimed at fiction writers. And with that, comes a lot of controversy. Sudowrite has been called " an insult to writers everywhere " and has been generally dismissed as a tool for hacks by a lot of Very Online writers. And while it's true that it's nowhere close to replacing a human author, it's fun, functional, and can genuinely help with writing a work of fiction. 

The Story Engine feature, which allows you to generate a full work of fiction over a few days by progressively generating each story beat, has attracted the most attention ( it works but takes lots of hand-holding and your novel will be weird ). But I prefer its assistive tools.

Let's start with Describe. Select a word or phrase, click Describe , and the AI will generate a few suggestions for the sight, smell, taste, sound, and touch of the thing, as well as a couple of metaphors. If you're the kind of writer who struggles to add sensory depth to your short stories, it can help you get into the habit of describing things in more interesting ways.

Then there's Brainstorm. It allows you to use the AI to generate possible dialogue options, character names and traits, plot points, places, and other details about your world from your descriptions and cues. If you know you want a big hairy guy with a huge sword but can't think of a good name, it can suggest a few, like Thorgrim and Bohart.

And these are just scratching the surface. Sure, if you over-rely on the AI to solve all your problems, you'll probably end up with an impressively generic story. But if you use it as a writing buddy to bounce ideas off and get you out of a rut, it's got serious potential. 

Best of all, Sudowrite is super easy to use. The onboarding, tool tips, and general helpful vibe of the app are something other developers could learn from. 

Sudowrite pricing: Hobby & Student plan from $19/month for 30,000 AI words/month. 

Best AI text generator for a non-GPT option

Writer (web).

Writer pros:

Not based on GPT, so free of a lot of the controversy surrounding LLMs

Surprisingly capable as an editor, making sure your team sticks to the style guide and doesn't make any wild claims

Writer cons:

Requires a lot more setup to get the most from

GPT comes with quite a lot of baggage. OpenAI has been less than transparent about exactly what data was used to create the various versions of GPT-3 and GPT-4, and it's facing various lawsuits over the use of copyrighted material in its training dataset. No one is really denying that protected materials— potentially from pirated databases —were used to train GPT; the question is just whether or not it falls under fair use. 

For most people, this is a nebulous situation filled with edge cases and gray areas. Realistically, it's going to be years before it's all sorted out, and even then, things will have moved on so far that the results of any lawsuit are likely to be redundant. But for businesses that want to use AI writing tools without controversy attached, GPT is a no-go—and will be for the foreseeable future. 

Which is where Writer comes in.

Feature-wise, Writer is much the same as any of my top picks. (Though creating a specific brand voice that's automatically used is an Enterprise-only feature; otherwise, you have to use a lot of checkboxes in the settings to set the tone.) Some features, like the chatbot, are a little less useful than they are in the GPT-powered apps, but really, they're not why you'd choose Writer.

Where it stands out is the transparency around its Palmyra LLM . For example, you can request and inspect a copy of its training dataset that's composed of data that is "distributed free of any copyright restrictions." Similarly, Palmyra's code and model weights (which determines its outputs) can be audited, it can be hosted on your own servers, and your data is kept secure and not used for training by default. As an AI-powered tool, it's as above board as it comes.

In addition to generating text, Writer can work as a company-specific Grammarly-like editor, keeping on top of legal compliance, ensuring you don't make any unsupported claims, and checking that everything matches your style guide—even when humans are writing the text. As someone who routinely has to follow style guides, this seems like an incredibly useful feature. I wasn't able to test it fully since I don't have a personal style guide to input, but Writer correctly fixed things based on all the rules that I set.

In side-by-side comparisons, Writer's text generations sometimes felt a little weaker than the ones from Jasper or Copy.ai, but I suspect a lot of that was down to how things were configured. Writer is designed as a tool for companies to set up and train with their own data, not run right out of the box. I'd guess my random blog posts were a poor test of how it should be used in the real world.

Writer also integrates with Zapier , so you can use Writer to create content directly from whatever apps you use most. Learn more about how to automate Writer , or take a look at these pre-made workflows.

Create new outlines or drafts in Writer based on briefs from Asana

Generate marketing content from project briefs in Trello

Writer pricing: Team from $18/user/month for up to 5 users; after that, it's an Enterprise plan.

Best AI text generator for GPT-4 content

Writesonic (web).

Writesonic pros:

Allows you to select what GPT model is used to generate text 

Generous free plan and affordable paid plans 

Writesonic cons:

A touch too focused on SEO content for my taste

While almost all the tools on this list use GPT, most are pretty vague about which particular version of it they use at any given time. This matters because the most basic version of the GPT-3.5 Turbo API costs $0.002/1K tokens (roughly 750 words), while GPT-4 starts at $0.06/1K tokens, and the most powerful version costs $0.12/1K tokens. All this suggests that most apps may not use GPT-4 in all circumstances, and instead probably rely on one of the more modest (though still great) GPT-3 models for most text generation. 

If having the latest and greatest AI model matters to you, Writesonic is the app for you. Writesonic doesn't hide what AI model it uses. It even allows you to choose between using GPT-3.5 and GPT-4, at least on Business plans. 

Whether the content you create will benefit from the extra power of GPT-4 or not depends. In my experience using GPT-4 through ChatGPT, the latest model is more accurate and, essentially, more sensible in how it responds. If you're churning out low-stakes copy variations for your product listings, you likely won't see much improvement. On the other hand, for long-form original blog posts, it could make a difference. Either way, the transparency in which model you're using at any given time is a huge bonus. 

Feature-wise, Writesonic is much the same as any of the other apps on this list, with a Google Docs-style editor, the option to set a brand voice, a few dozen copy templates, a chatbot, a browser extension, and Surfer integration. It's cool that you can set reference articles when you're generating a blog post, but it introduces the real possibility of inadvertent plagiarism if you aren't careful with how you use it. (Its most offbeat feature is a surprisingly solid AI-powered custom chatbot builder that's due to be spun out into its own app soon.) Overall, it's pretty nice to use and skews more toward SEO-optimized content marketing—but like with all the apps, you can use it to generate whatever you want.

Writesonic also integrates with Zapier , so you can send new copy to any of the other apps you use in your writing workflow. Learn more about how to automate Writesonic , or get started with one of these examples.

Create a Google Doc with new content from Writesonic

Generate product descriptions with Writesonic from spreadsheet rows in Google Sheets

Writesonic pricing: Free for 10,000 GPT-3.5 words per month; Business from $19/month for 200,000 Premium words or 33,333 GPT-4 words.

Best free AI writing generator (with affordable upgrades)

A solid free plan and a cheap high-volume plan (though Writesonic offers better value for an unlimited plan)

It includes a basic AI art generator as part of every plan 

The app is more basic than more expensive offerings

Unlimited plan isn't very competitive

Most of the apps on this list are aimed at professionals, businesses, and anyone else with a budget. The Jasper, Copy.ai, and Anyword plans I considered all started at $49/month. That isn't exactly a hobbyist-friendly sum of money, so if you want to explore AI text generators without spending as much, give Rytr a go.

There's a free plan that's good for 10,000 characters (around 2,500 words) per month, and it includes a lot of the features, like a plagiarism checker, and a few AI-generated images. The Saver plan starts at $9/month and allows you to generate 100,000 characters (around 25,000 words) per month. On that plan, you're also able to generate up to 20 images a month, which many other apps charge extra for. (There's also an unlimited plan for $29/month, but at that point, Writesonic is a better value.)

Feature-wise, there are some trade-offs. Rytr is a little less competent at generating long-form content without you guiding it through the process, and there are fewer templates for specific things. The interface also isn't as polished, and there isn't as much hand-holding to get you started. Still, as Rytr is using GPT like almost all the other apps on this list, you should be able to get it to produce substantially similar output.

Rytr Pricing: Free plan for 10,000 characters/month and lots of other features; Saver plan from $9/month for 100,000 characters; Unlimited plan from $29/month.

Other AI writing tools to consider

With so many AI text-generating tools out there, a few good ones worth considering didn't make this list, only because they didn't meet my initial criteria in some way. If none of the AI writers I chose fit the bill for you, here are a few other options worth looking into:

ChatGPT is surprisingly competent and fun to use. And best of all, it's free. ( Google Bard is a little less excellent on the content production side.) 

Wordtune and Grammarly are both great tools for editing and improving your own writing .  GrammarlyGO just isn't as flexible as my other picks.

Notion AI adds a powerful AI tool directly into Notion. If you already use Notion, it's worth checking out, but it's a lot to learn if you just want a text generator. (Same goes for AI within any other Notion alternative, like Coda AI .)

Surfer and Frase are both AI-powered SEO tools. They fell slightly out of scope for this list, but they can both help you optimize and improve your content—AI-generated or not. 

All of the apps on this list offer at the very least a free trial, so I'd suggest trying some of them out for a few minutes until you find the one that seems to work best with your workflow.

Related reading:

How to use OpenAI's GPT to spark content ideas

How to create an AI writing coach with GPT and Zapier

8 ways real businesses are using AI for content creation

How to detect AI-generated content

This article was originally published in April 2023. The most recent update was in September 2023.

Get productivity tips delivered straight to your inbox

We’ll email you 1-3 times per week—and never share your information.

Harry Guinness

Harry Guinness is a writer and photographer from Dublin, Ireland. His writing has appeared in the New York Times, Lifehacker, the Irish Examiner, and How-To Geek. His photos have been published on hundreds of sites—mostly without his permission.

• Artificial intelligence (AI)
• Media and editorial

Related articles

The 8 best meeting scheduler apps in 2023

The 21 best sales tools in 2023

The best AI email assistants in 2023

The 7 best no-code app builders in 2023

Improve your productivity automatically. Use Zapier to get your apps working together.

Should AI Write Research Papers on Behalf of Authors?

Artificial Intelligence (AI) has made it to the field of research publishing . In the past decade, AI and machine learning have transformed several industries. The disruptive technology of AI is making it easier and faster to automate several processes. In the academic publishing industry, AI-based technologies are being developed and implemented to assist both authors and publishers. These technologies help tackling issues related to peer review, searching published content, detecting plagiarism, and identifying data fabrication. Thus, AI can not only help expedite scientific communication but also reduce human bias.

Similarly, the application of AI in research has grown tremendously with a focus on automation of research techniques from generating a hypothesis to conducting experiments. In fact, researchers are now being able to address complex problems in biomedical sciences, drug combinations, and predicting diseases using AI.

AI Tool for Writing Manuscripts

Among the several AI-based tools, we now have a tool that can write manuscripts. Manuscript Writer, a tool launched by SciNote, has the ability to significantly simplify the process of preparing scientific manuscripts by using the technological advances in machine learning and artificial intelligence. Recognizing the importance of timely publication of scientific findings by the global scientific community, the tool aims to significantly reduce the time taken to prepare initial content . The tool draws upon data contained within the ELN of SciNote and references that are accessible from open access journals, to prepare a structured draft. This draft is then shared with the author to edit and develop further.

How Will AI Avoid Plagiarism in these Research Papers?

Manuscript Writer pulls data from the selected references, and based on the relevant keywords it will look for additional relevant open access references and include them in the draft as well. After every paragraph in the introduction, the number of the reference and a percentage pops up to the scientist. This shows the scientist that a particular paragraph is cited from the specified reference and is X% the same text. This is a clear indicator for the scientist to edit and make the necessary changes in the manuscript as they would have done while writing their manuscript. Therefore, whether the text stays the same or not is the responsibility of a scientist. Moreover, a ‘Manuscript Writer’ cannot take over their own responsibility of not editing the text.

The main benefit of using Manuscript Writer is the inclusion of interesting paragraphs, related to the subject at hand, to the introduction and give the scientist a head start while writing. Manuscript Writer’s purpose is not to write the finalized text instead of the scientist, its purpose is to empower the scientist. Which is why it cannot write the discussion section, which is the most creative and original part of the scientific article and greatly depends on the scientist’s style and way of thinking. Every scientist adds their own expertise and knowledge to the entire text.

Should AI Write Research Papers for Them: What do The Researchers Say?

Researchers, however, feel divided when it comes to answering this question. As mentioned previously, the draft will be generated not just from the data stored by the user. It will collect data from from “relevant keywords and open access references.” Obviously, an AI isn’t capable of understanding the prose the way a human might. So it is difficult to predict how it is going to create any sort of derivative work based on open-access references that isn’t plagiaristic or incoherent (or most likely both.) In fact several researchers compare it with the SCIgen tool that came into the news for producing fake scientific papers.

Nevertheless, researchers and cofounders seem to be optimistic about the AI tools. AI tools have been useful previously as search engines. For example, Yewno, one such AI-based search engine tool for Education. It allows researchers to obtain multidisciplinary content in the areas of medicine, biology, physical sciences, and many more. In fact, they plan to add financial services as one of the disciplines in the near future. Owen Etzioni, chief executive at AI2, aims to have the Semantic Scholar system index all of PubMed and expand in the medical sciences. AI has made its use in the process of peer review. Hence we can be hopeful of the new AI tools like Manuscript Writer for improving and progressing scholarly publishing.

Rate this article Cancel Reply

Your email address will not be published.

Enago Academy's Most Popular

• AI in Academia

Maintaining the Ethical Boundaries of Generative AI in Research and Scholarly Writing

Generative AI models have revolutionized the way we interact with technology for research and scholarly…

• Reporting Research

Preserving Research Integrity: Why author guidelines on generative AI tools matter

After COPE, the Committee on Publication Ethics, along with other heavyweights like WAME (World Association…

• Manuscripts & Grants
• Trending Now

Call for Responsible Use of Generative AI in Academia: The battle between rising popularity and persistent concerns

Generative AI is a fascinating and complex landscape, but it is not without its inherent…

Limitations of DeepL Write and ChatGPT in Scholarly Editing: How human editors overpower AI tools

Complete reliance on AI, such as DeepL Write and ChatGPT, for research and scholarly editing…

• Career Corner
• PhDs & Postdocs

Avoiding the AI Trap: Pitfalls of relying on ChatGPT for PhD applications

The PhD application process can be both daunting and enjoyable, where one tends to make…

The Future of Research Writing: Top universities, publishers, and policymakers share…

Can AI-Based Search Tools Replace Google Scholar?

Sign-up to read more

Subscribe for free to get unrestricted access to all our resources on research writing and academic publishing including:

• 2000+ blog articles
• 50+ Webinars
• 10+ Expert podcasts
• 50+ Infographics
• 10+ Checklists
• Research Guides

We hate spam too. We promise to protect your privacy and never spam you.

I am looking for Editing/ Proofreading services for my manuscript Tentative date of next journal submission:

What are your major challenges while writing a manuscript?

Culture and Growth

Using Artificial Intelligence (AI) to Write Research Papers

This article was written partly by chatgpt.

The scholarly communication industry is investing heavily in writing tools to help researchers more quickly draft and submit research articles. Why?

It is estimated that global spending on scientific research and development (R&D) was around $1.7 trillion in 2020 and has continued to grow. Most of that is spent on salaries, lab spaces, equipment, and supplies. But a material amount is also spent in publishing. According to some estimates, the global market for scholarly publishing was worth around $30 billion. It flies under the radar in discussions of media, where music and television take the headlines.

How AI Can Help

As we develop AI tools and assistants, we can reduce the burden on researchers to spend time writing and submitting articles, freeing them to do more research. That’s where writing tools come in. AI technology can be used to help write research papers in a number of ways. For example, AI algorithms can be trained to automatically generate research papers based on data and information provided by the user. These algorithms can analyze large amounts of data and information, and can use natural language processing (NLP) to generate high-quality research papers in a fraction of the time it would take a human.

Additionally, AI can be used to assist with the editing and proofreading of research papers, like in AJE AI Editing , Writefull , and Trinka AI . AI algorithms can be trained to automatically identify and correct errors in grammar, spelling, and punctuation, and can provide suggestions for improving the overall clarity and coherence of the paper. This can help researchers save time and improve the quality of their research papers.

Furthermore, AI can be used to help with the research and data analysis process. AI algorithms can be trained to analyze large amounts of data and information, and can help researchers identify patterns and trends that may not be immediately apparent. This can help researchers generate new insights and ideas for their research papers.

In conclusion, AI technology offers many potential benefits for researchers and can help improve the efficiency and effectiveness of the research paper writing process. By leveraging the power of AI algorithms, researchers can save time and effort, and can produce high-quality research papers that are more likely to be accepted by academic journals. Humanity is better served by reducing the time needed by researchers in the writing and editing process, freeing more of their precious time to advance human knowledge.

This article was written partly by ChatGPT, which can be found here:

https://chat.openai.com/chat

So which parts of this article were written by ChatGPT? I added the opening to set out the context, added the sepcific examples, and wrote the ending. The meat of the article was written by ChatGPT.

Leave a comment

Thank you for reading Culture and Growth. This post is public so feel free to share it.

Ready for more?

Paperpal for Researchers Join the academic and scientific writing revolution

Create impactful manuscripts and fast-track journal submissions with our smart writing tools for researchers

Showcase your research with high-quality academic writing

The academic writing revolution is here; with the emergence of new research paper writing tools, content creation has never been easier. Whether your objective is tenure, a research grant, a book deal, or just an improved reputation in your department, a string of high-profile journal articles are the building blocks you need to fast-track your academic career. As a researcher, what better way of communicating your research than by showcasing it in a well-written article that is published in a top journal? However, the process of journal submission and multiple research paper grammar checks to ensure the language is of the highest quality isn’t without its challenges. A global Editage study in 2018 found that nearly half of the authors surveyed faced challenges in preparing a manuscript for journal submission and found peer review to be a daunting process. Paperpal, with its AI-powered tools for researchers, is the key to making this process a simpler, faster one for authors everywhere.

Enhance your academic writing skills from the first draft itself.

Paperpal is the perfect tool for researchers at every writing stage

Enhances and speeds up the academic writing process.

There’s more to academic writing than simply knowing what to write. But delivering a strong, well-written manuscript is not easy and even ground-breaking research papers risk desk rejection because they are hard to follow or because of avoidable errors like poor grammar, spelling, and punctuation. By the time you finish the first draft, you would have already invested a significant amount of time and effort in communicating your findings accurately. This is then followed by multiple rounds of editing and research paper grammar checks to refine your work for submission, which can delay the publication of possibly time-sensitive results. Although academic writing isn’t easy, scientific writing tools for researchers powered by artificial intelligence and machine learning are transforming the experience. Paperpal for Word for instance provides real-time suggestions to improve your grammar, spelling, punctuation, clarity and structure, empowering you to enhance and speed up the writing process from the first draft itself.

Simplifies and optimizes scientific editing for researchers

It’s not just academic writing, but high-quality English language editing plays a key role in improving your writing style and ensuring clarity. Finding an online tool that acts as a basic spelling, grammar, and sentence corrector is straightforward, but finding thesis writing and scientific editing tools that are tailored for researchers and academic writing is not so easy. There is a need for precise scientific editing, which includes accurate research paper grammar checks, using the right language, understanding and applying the necessary academic writing conventions, and getting the style and structure right. This is where Paperpal can help. In a 2022 University of Cambridge study of seven English editing tools for researchers, Paperpal stood out as the preferred English editing and scientific writing tool for researchers. It suggested a high number of accepted edits, providing alternative words and phrases that were in line with those made by human editors to enhance the language and readability of text. Paperpal was also the simpler tool for researchers to use for editing and research paper grammar checks.

Streamlines journal submissions with comprehensive checks

Imagine if after all the work you put in preparing your manuscript for submission, you’re rejected because it fails the basic technical checks. Being rejected for avoidable snags like a missing conflict of interest statement or ethics statement is not only disappointing, it further delays research publication. Most journals have a core set of submission requirements that need to be followed in order for a manuscript to be considered. But with so many things to check and do, researchers often stumble at this stage. This is where Paperpal for Manuscript, which checks your research paper across key language and technical parameters, is the perfect tool for researchers to check their submission readiness. Upload your ready manuscript and for just $29 you can download a Word file with all the suggestions included in mark-ups or comments. Accept or reject suggestions with a few clicks, and recheck your work unlimited times to create the best version of your manuscript. Premium editing has never been simpler.

Ready to create powerful manuscripts that will impress journal editors?

Paperpal for Word

Paperpal for Web

Use Paperpal both online and offline

Paperpal for Word is a thesis writing and scientific editing tool for researchers who want to polish their academic writing as they work. The plugin is easy to install and use, making it a great tool for researchers to consistently deliver high-quality academic writing. If you don’t want to commit to the Word add-in, you can explore Paperpal for Web, a robust research paper writing tool that has an in-built grammar checker for scientific writing. All you need to do is write, paste or even upload your academic text into your web browser to get instant suggestions on how to fix your language and grammar. Once you’re ready to submit, take this up a notch with Paperpal for Manuscript. With instant checks for disclosures, figures and/or tables, word counts, references, language, structure, and much more, this secure and trusted academic research tool should be part of every researcher’s pre-submission kit.

Get the premium editing your paper needs and deserves.

Levelling the playing field for researchers

English has long been established as the language of academic publications. And even then, different journals prefer a specific variation, such as British or American English. While this can trip up even native English-speaking researchers, it makes the task of creating a compelling high-quality manuscript for publication even more challenging for those with English as a second language. Paperpal’s AI-powered research paper grammar check and scientific writing tools were built with a vision to democratize scholarly publishing by ensuring your academic writing meets the highest publication standards. Where you are in the world doesn’t matter. We’re part of an academic writing revolution that empowers anyone, anywhere to polish their academic writing skills with access to our precise research paper writing tools for researchers. Paperpal differentiates between British and American English and offers appropriate suggestions in terms of spellings, vocabulary, pronunciation, and grammar. Moreover, our AI is trained on millions of corrections made by professional editors across 1,300 subject areas, which means you get tailored suggestions to enhance your research writing and boost your chance of publication success.

Amplify your reach, impact, and researcher reputation by delivering high-quality manuscripts.

Trusted by top global publishers and academic societies.

Paperpal is the preferred AI writing assistant for more than 20,000 academics and endorsed by 13 leading publishers with over 400 journals across the globe. Journals are now offering the Paperpal Preflight tool for researchers on their websites, which allows you to upload your manuscript and check it against the journal’s house style before submission, reducing the risk of desk rejection. Our partners include Wolters Kluwer, Cambridge University Press, and The American Association for Cancer Research, among others and this list is only set to grow as we partner with the best. See our full list of publisher partners below

Want practical strategies and expert advice on writing, editing, and submission?

©2002- 2023 Cactus Communications.

All Rights Reserved.

Your documents are always safe

ISO/IEC 27001- 2013 CERTIFIED

Cactus Communications

Services Pte Ltd

20 McCallum Street, #19-01,

Tokio Marine Centre,

Singapore 069046

Free All-in-One Office Suite with PDF Editor

Edit Word, Excel, and PPT for FREE.

Read, edit, and convert PDFs with the powerful PDF toolkit.

Microsoft-like interface, easy to use.

Windows • MacOS • Linux • iOS • Android

• Create with AI
• WPS AI Tutorial

How to Use WPS AI/Chatgpt to Write Research Papers: Guide for Beginners

Many students are unaware of the incredible potential AI holds to revolutionize their work. By harnessing the power of AI, they can enhance the quality and efficiency of their research, ensuring a higher grade and solidifying their academic credibility. In this article, we will explore numerous ways you can leverage the two powerful AI tools for writing research papers: WPS AI and ChatGPT to learn practical prompts that can be customized to suit individual needs.

Part 1. How to Use WPS AI to Write Research Papers

Among the Writing AI tools, WPS AI emerges as a revolutionary tool specifically designed to enhance the research paper writing experience.

WPS AI empowers researchers to delve into their subjects with greater depth and efficiency, catalyzing productivity and enabling the pursuit of scholarly excellence

How to Download WPS AI

Step 1: Visit WPS AI on your browser

Step 2: Users must sign-up to gain early access to WPS AI.

Step 3: Users will be asked to fill out the registration form to register for early access.

Step 4: Once registered, users will be sent a guide on how to download and install WPS AI on their PC through an email.

Step 5: Users can launch WPS AI on their systems once WPS AI is installed.

How to Use WPS AI to Write Research Papers

WPS AI, with its user-friendly interface and other advanced functionalities makes the process of writing research papers a piece of cake.

The following methods are ways in which WPS AI can assist you in your research paper writing:

Step 1: Open WPS Office to create a new document in WPS Docs.

Step 2: Click on the "WPS AI" in the ribbon menu or simply type "@AI".

Step 3: In the prompt box, type "write a paper on {enter title of research paper}” OR “Write a research paper about".

Step 4: You can alter the response or change the question once it has been generated. To do this, select "Return" from the response's menu.

Step 5: Users can simply Continue/Rewrite/ Accept/Discard the results

Use Word, Excel, and PPT for FREE, No Ads.

Edit PDF files with the powerful PDF toolkit.

Microsoft-like interface. Easy to learn. 100% Compatibility.

Boost your productivity with WPS's abundant free Word, Excel, PPT, and CV templates.

Part 2. How to Use ChatGPT to Write Research Papers

Writing research papers can be time-consuming, involving tasks like summarizing findings and describing methodologies. But now, researchers have a powerful ally in ChatGPT. By entrusting repetitive tasks to ChatGPT, users can streamline their workflow, saving valuable time. This allows researchers to focus more on enhancing their papers and conducting core research, leading to stronger and more insightful results.

With the help of ChatGPT, researchers can create more compelling title for their research papers

Step 1: Open ChatGPT in your browser.

Step 2: In the ChatGPT chatbox, type in a prompt asking ChatGPT to create a title for your research paper using your paper’s abstract

Researchers can also ChatGPT to write an abstract by providing the title of the research paper

Step 1: Give ChatGPT a prompt asking it to write an abstract based on the title. ChatGPT will craft an abstract following the correct formatting and pattern of an abstract.

Similarly, researchers can utilize ChatGPT as an AI assistant while writing research papers for Research gaps, research questions, Literature review.

Step 1: Ask ChatGPT to write a research gap with help of your Research paper title

Step 2: To write better research questions, give a prompt to ChatGPT to write research questions using your research gap. Chatgpt research questions are very specific and follow the proper format.

Step 3: By providing your topic and asking for key publication, researchers can eliminate the process of manually searching for key publication.

This is how researchers can make use of ChatGPT to help them out craft insightful research papers.

"Please provide a summary of the key findings and conclusions of this research paper."

This prompt asks the AI to generate a concise summary highlighting the main findings and conclusions of the research paper.

"Can you give a brief overview of the methodology used in this research paper?"

This prompt focuses on the research methodology employed in the paper, asking the AI to provide a summary of the approach or methods used.

"Summarize the main contributions and significance of this research paper."

This prompt seeks to extract the primary contributions and significance of the research paper, capturing its unique value or impact in the field.

"Provide a summary of the literature review presented in this research paper."

This prompt directs the AI to summarize the literature review section of the paper, which highlights the existing knowledge and previous studies related to the research topic.

"Can you briefly outline the experimental design and results of this research paper?"

This prompt requests a summary of the experimental design or methodology along with a concise summary of the research results obtained.

"Summarize the theoretical framework and hypotheses explored in this research paper."

This prompt focuses on the theoretical framework and hypotheses put forth in the paper, asking the AI to provide a summary of these foundational elements.

"Summarize the limitations and future directions discussed in this research paper."

This prompt directs the AI to summarize the limitations or constraints of the research study and any potential avenues for future research identified by the authors.

"Summarize the implications and practical applications of the research findings presented in this paper."

This prompt asks the AI to highlight the practical implications and real-world applications of the research findings discussed in the paper.

Q1. Will ChatGPT replace technical writing?

ChatGPT is an AI tool that can be used to streamline and enhance certain aspects of technical writing, however, it is unlikely to replace humans. Technical writing involves complex and specialized knowledge, which ChatGPT currently lacks. It can help users with structures, provide a format for technical writing and various other aspects, but the information provided could be factually wrong. Furthermore, ChatGPT produced content misses the human touch, which is vital to connect with your readers.

Q2. Does Google penalize AI written content?

No, Google does not penalize AI-generated content. Google has stated that AI material adheres to its policies. Google wants its users to get information and high quality content, and it wants content producers; websites, to provide this quality to them.

Q3. How can I effectively utilize ChatGPT for technical writing?

Users can utilize ChatGPT for technical writing by providing clear and detailed prompts. This would allow ChatGPT to produce a technical document with utmost details. Any missing information or errors can be fixed by providing further prompts asking ChatGPT to enhance and fix the document. ChatGPT can help the users to proofread their documents, provide guidelines on how to write a technical document, provide useful information, summarize long text, along with many other useful tasks.

Q4. Can ChatGPT help with formatting technical documentation?

Yes, ChatGPT can be very useful while formatting technical documentation. It can help in areas such as structuring sections, headers, and lists. However, because it may be unfamiliar with industry-specific formatting guidelines or style manuals, human input is required to ensure proper and comprehensive formatting.

Transform Your Research Papers with WPS AI

With the advent of AI, the once laborious task of working on research papers has now become remarkably streamlined, allowing for seamless progress and alleviating the burden of tedious efforts. While the market is brimming with AI solutions, one particularly promising tool emerges: WPS AI. Despite being in its beta version, it is predicted to make substantial strides, particularly in the realm of research. So integrate WPS AI with your WPS Office Suite today and discover more on how to use wps ai/chatgpt to+Research papers.

• 1. How to Use Wps AI/Chatgpt to Generate Charts A Comprehensive Guide for Beginners
• 2. How to Use ChatGPT for Beginners?-Register, Use Cases and FAQ
• 3. How to Use Wps AI/Chatgpt to Data Analysis: Guide for Beginners
• 4. How to Use Wps AI/Chatgpt to Write Technical Document: Guide for Beginners
• 5. How to Use Wps AI/Chatgpt to Summarize Meeting Notes: Guide for Beginners
• 6. How to Use WPS AI/Chatgpt to Write Essays: Guide for Beginners

15 years of office industry experience, tech lover and copywriter. Follow me for product reviews, comparisons, and recommendations for new apps and software.

Something went wrong. Wait a moment and try again.

Covid 19 - Lockdown, let out your PhD indagation with our Expert

• Live Support
• Quick Query
• [email protected]

Talk to our Consultant

Live Support in 214 Countries

Artificial intelligence research paper, get unique artificial intelligence research paper writing.

Artificial intelligence is a computer science field that develops algorithms, software, methods, and give computers the ability to think like a human. If you want to use AI to write essays and artificial intelligence in a research paper, you will have to write the topic in deep, do research and come out with a real verdict of your research so that it helps out readers.  The artificial intelligence (AI) field help us in publishing papers with a novel ideal and solving a problem.

Start with selecting a topic of your artificial intelligence research paper , find good references and sources and put it all in the paper. You can always seek a professional's help who will guide you better and help you in preparing the most authentic and unique paper. Words Doctorate have qualified artificial intelligence subject writers on board who welcomes basic research and prepare it in mature, complete, and articulate research paper on artificial intelligence.   

Order Research Paper Writing in AI (Artificial Intelligence)

Using AI professional writers to write papers saves time that you can use your time to do another important task like viva preparation, exam preparation, or part-time job.  Our writers and editors can manage your paper writing and you won’t have to worry about how they’ll turn out.

Our writers focus only one client at a time so they can give 100% attention to the paper writing work. One of the benefits of letting our writers write research papers for you is that you can lead a stress-free life and get the best paper delivered quickly.

To take the stress and hours of work out of your research paper, you can order research paper writing in artificial intelligence and then relax knowing that your paper is being completed by the pros.

Hire Artificial Intelligence Research Paper Writer

Our expert writers at Words Doctorate have written more than 200+ papers for researchers around the world in Artificial intelligence subject. You will be assured to get unique 100% plagiarism free paper and 24/7 quick support by professionals.

We will handle the customer's request at our top priority and ensure that you get the best papers on your requirement. Contact us and hire artificial intelligence research paper writer to help you with best paper work.

Send Us Message

Running projects.

Computer Science Dissertation Writing Services

Social Science

Biotechnology - Biology

Civil Engineering

Medical-Pharmaceutical

Management Dissertation Writing Services

Electrical – Electronics

Agriculture

• TechRepublic

Account Information

Microsoft Copilot Cheat Sheet: Complete Guide for 2023

Share with Your Friends

Your email has been sent

This comprehensive guide about Microsoft Copilot includes details about the AI product's benefits, pricing, functional integrations and alternatives.

The practical application of generative artificial intelligence has gone from an abstract, future concept to a concrete reality in a matter of mere months. Businesses and organizations large and small are scrambling to figure out if and how AI can help their people be more productive and efficient. For organizations using Microsoft software, the application of AI in a business environment is being led by the Microsoft Copilot platform.

Businesses at the enterprise level are also looking for ways AI can leverage the massive amounts of data generated daily by their organizations in increasingly productive ways. Many believe that such a massive undertaking can only be accomplished by a competent AI platform. Microsoft Copilot, with its integration into Microsoft 365, Azure, Windows and enterprise-wide data streams, is purported to be the AI that unlocks the creative and productive potential of an organization’s people and data.

What is Microsoft Copilot?

What are business benefits to using microsoft copilot, how does microsoft plan to integrate copilot into its applications, what are potential problems and caveats with microsoft copilot, what are the alternatives to microsoft copilot, how much does microsoft copilot cost, which businesses should consider microsoft copilot, and which should not, when will microsoft copilot be available.

Microsoft Copilot is a new AI product that combines the power of large language models with in-house enterprise data generated by the Microsoft Graph and Microsoft 365 applications. Using the power of AI and natural language conversations, users can find better answers to their questions and potentially create content from those answers. Copilot was developed on the ChatGPT platform and announced as an in-development platform at the July 2023 Microsoft Inspire conference .

There are two versions of Copilot: Microsoft 365 Copilot and a more general Microsoft Copilot. It is important to note that Microsoft 365 Copilot is different from the consumer and small business-oriented Copilot platform found in Bing Chat or non-enterprise versions of Microsoft 365. These Copilot LLMs are trained on more generalized aggregate data gathered across the internet and therefore tend to have more generalized results. Microsoft 365 Copilot will be dependent on the data generated by a specific, and only a specific, enterprise.

Microsoft 365 Copilot will be dependent on in-house, enterprise-generated data, while the more general Microsoft Copilot will use aggregate data pulled from the internet.

Both Copilot versions will be embedded in the Microsoft 365 apps including Word, Excel, PowerPoint, Outlook and Teams. This complete Microsoft 365 integration will allow an organization’s workers to be more creative and unlock productivity gains and potentially improve their skills. In addition, Microsoft 365 Copilot will add AI-enabled Business Chat to the productivity suite, which will work across enterprise-specific data like calendars, emails, chats, documents, meetings and contacts to help employees communicate easier and better.

Assuming Microsoft Copilot works as advertised, an employee could jump-start a project (e.g., email, presentation, report, data visualization) with an AI-generated first draft. With that foundation, the employee can quickly move on to refining and iterating a second draft and then a final draft, cutting at least one step from the process. This should make the employee more efficient and productive, and it could allow for the development of new skills.

For Microsoft 365 Copilot, the key to these productivity gains will be the application of enterprise-specific data to the new project from the start. Using data internally generated by the enterprise from emails, documents, calendars, contacts and so on, presumably the project will be jump-started with limited, and more importantly, pertinent assets.

SEE: Hiring kit: Prompt engineer (TechRepublic Premium)

For example, if an enterprise employee at Ford is creating a presentation that needs to show movement, the AI – if properly trained – should use a Ford-related product rather than a similar product from one of its competitors. An AI trained by data culled from the internet may come to a different conclusion and defeat the purpose of deploying enterprise-specific generative AI.

Copilot will be integrated into the fabric of all Microsoft 365 applications. When an employee starts a Word document, reads an email in Outlook, opens an Excel report or updates a PowerPoint presentation, Copilot and its generative AI abilities will be there to assist when called upon.

The specifics of how each Microsoft 365 application will use Copilot are still in development, and many use cases will likely only be discovered when users can actually use the platform. However, Microsoft has outlined some of its basic ideas for Copilot in a business environment .

Below are examples of commands a user might give Microsoft Copilot.

Copilot in Word

• Draft a two-page project proposal based on data gleaned from a Word document (i.e., either the one you are currently working on or one that you specify by name) and an Excel worksheet.
• Make the third paragraph in the current document more concise and change the tone of the document to be more casual.
• Create a one-page draft based on this rough outline.

Copilot in Excel

• Break down this sales data by type and channel and then insert a table.
• Project the impact of a variable change in this data and then generate a chart to help visualize it.
• Model how a change to the growth rate for a variable would impact my gross margin.

Copilot in PowerPoint

• Create a five-slide presentation based on this Word document and include relevant stock photos.
• Consolidate this presentation into a three-slide summary.
• Reformat these three bullets (in a specific PowerPoint presentation) into three columns, each with a picture.

Copilot in Outlook

• Summarize the emails missed while out of the office last week and flag any important items.
• Draft a response thanking the senders of an email and asking for more details about their second and third points.
• Shorten a draft email and make the tone more professional.
• Invite everyone to a lunch-and-learn about new product launches next Thursday at noon. Mention that lunch is provided.

Copilot in Business Chat

• Summarize the chats, emails and documents about the topic being discussed.
• What is the next milestone for a project? What risks were identified? Brainstorm a list of potential mitigations.
• Write a new planning overview in the style of this (specified existing) document that contains the planning timeline from a different document and incorporate the project list in the email from this person (specifying a user’s name).

All AI platforms are only as good as their training; if Microsoft Copilot is modeled after data that is incomplete, biased, wrong or otherwise corrupt, the suggestions it generates, regardless of who is asking, will be incomplete, biased, wrong or otherwise corrupt. The old adage of garbage in equals garbage out still applies.

For Microsoft 365 Copilot, training will be particularly important because all the data used for that Copilot AI platform will be internally generated and gathered. Enterprises looking to properly apply AI will have to carefully monitor the data Copilot can access. For example, a business may not want brainstorming documents and meeting notes to be part of the data stream; rejected ideas may taint the data stream and propagate throughout the organization.

For general versions of Microsoft Copilot, organizations and users will have to be aware that inherent biases, fads, misguided trends and other transitory events will likely color some of the AI-generated output. Not every passing fancy on the internet should make its way into official organizational documents.

In addition to this potential problem, especially for Microsoft 365 Copilot implementations, is the tendency of some departments to silo their data behind firewalls. New products and services developing under non-disclosure agreements, for example, will often be cut off from the normal organizational data stream. Businesses will have to decide whether hiding data from the AI in such cases is more beneficial than allowing AI access.

The most important caveat for Microsoft Copilot is the platform is still in development. The use cases outlined by Microsoft at the 2023 Inspire conference are the company’s visions of how the platform will work. Once Copilot is released to the public, what the AI platform is capable of, good or bad, will be revealed. Until then, we will be dealing with potential and perhaps some wishful marketing.

Generative AI is arguably the hottest trend in technology innovation for 2023, so it stands to reason there are many new and in-development AI platforms ready to compete with Microsoft Copilot.

Even though Copilot is based on ChatGPT, the AI chatbot is available as a standalone platform and therefore should be considered a competitor. AI is already incorporated into Microsoft Edge in the form of Bing Chat, and an AI has recently been released for the Google search engine and Chrome . Technology experts have reported that Apple is developing its own AI platform. It seems that AI will be integrated into just about every digital application we use on a computing device.

Major tech companies including Salesforce , Oracle and Adobe are all working on AI platforms. There are also dozens of smaller independent developers working on their own versions of an AI platform. Plus, there are a multitude of AI competitors working on specialized platforms that will bind LLMs and generative AI principles to specific applications. Businesses will likely spend a lot of time wading through AI platform possibilities.

As revealed during the 2023 Inspire conference, Microsoft 365 Copilot will cost $30/user/month. At first glance, this price point seems expensive, but it is vital to remember this version of the platform is designed for large business enterprises. For a large enterprise with thousands of employees, and assuming the platform delivers what Microsoft promises, that $30 could end up being a bargain. That is a big assumption to make at this early point in development.

The consumer and SMB versions of Microsoft Copilot will likely be priced lower than the enterprise counterpart. Bing Chat , which is also based on ChatGPT, is available now and for free as an integral part of Microsoft Edge. It is also likely that some features restricted or otherwise modified version of Copilot will be available for SMBs too small to have meaningful in-house generated data available to train the AI platform.

The pricing for all the versions of Microsoft Copilot is likely to change as development of the platform continues. With the large number of competitors in the AI platform space, it seems almost certain the cost of these services will change significantly.

At $30/user/month, only certain large enterprises will be able to afford large numbers of employees subscribing to the Microsoft 365 Copilot platform. In addition to the subscription fee, such employers will also have to account for the extra expenses associated with generating, collecting and collating accurate and useful data for the AI and LLMs to train on. This is a major undertaking, and the decision to implement the Copilot platform will take a significant commitment.

Large enterprises will have to decide whether the productivity benefits of using the Microsoft 365 Copilot platform outweigh the initial costs of developing and maintaining the platform – and then paying for it indefinitely.

For individuals and small businesses with little to no LLM-ready data, the consumer level versions of Microsoft Copilot is available for free. Bing Chat is already available in Microsoft Edge for everyone. Some form of low-cost or no-cost version of Copilot will likely be available for certain versions of Microsoft 365.

Only time will tell if the generative AI capabilities of Copilot are worth the time necessary to use them.

As of August 2023, Microsoft Copilot is in the testing phase of development. A limited number of organizations and Microsoft Insiders are testing the AI and providing feedback on what works and what does not work.

There is currently no officially announced release date for any version of Microsoft Copilot. Visit Microsoft Copilot

Subscribe to the Microsoft Weekly Newsletter

Be your company's Microsoft insider by reading these Windows and Office tips, tricks, and cheat sheets.

• Microsoft Build 2023: Microsoft extends its Copilots with open standard plugins (TechRepublic)
• How to hide the Discover button and Copilot in Microsoft Edge (TechRepublic)
• Microsoft adds Copilot AI productivity bot to 365 suite (TechRepublic)
• Microsoft’s First Generative AI Certificate Is Available for Free (TechRepublic)
• Artificial Intelligence: More must-read coverage (TechRepublic on Flipboard)

Contact Mark W. Kaelin

Your message has been sent

• See all of Mark W.'s content
• Artificial Intelligence
• Cheat Sheets

Editor's Picks

TechRepublic Premium Editorial Calendar: Policies, Checklists, Hiring Kits and Research for Download

TechRepublic Premium content helps you solve your toughest IT issues and jump-start your career or next project.

Microsoft’s First Generative AI Certificate Is Available for Free

Microsoft is also running a grant competition for ideas on using AI training in community building.

How Generative AI is a Game Changer for Cloud Security

Generative AI will be a game changer in cloud security, especially in common pain points like preventing threats, reducing toil from repetitive tasks, and bridging the cybersecurity talent gap.

The 8 Best International Payroll Services for 2023

Does your business need a payroll provider that offers international payroll services? Use our buyer's guide to review the best solutions, from ADP to Oyster.

ChatGPT Cheat Sheet: Complete Guide for 2023

Get up and running with ChatGPT with this comprehensive cheat sheet. Learn everything from how to sign up for free to enterprise use cases, and start using ChatGPT quickly and effectively.

6 Best monday.com Competitors and Alternatives for 2023

Looking for an alternative to monday.com? Our comprehensive list covers the best monday alternatives, their key features, pricing, pros, cons and more.

Hardware Inventory Policy

An organization’s IT hardware inventory constitutes a significant capital investment in order to perform company operations. Because hardware represents a critical asset and often holds organizational data, it is important for the company to track each component from purchase and deployment to decommissioning and disposal. This policy from TechRepublic Premium describes the guidelines by which ...

Vendor Comparison: DIY Home Security Systems

When it comes to DIY home security, there are many systems and components to consider and many decisions to make. This vendor comparison guide from TechRepublic Premium provides advice you can follow as you make decisions regarding how you will deploy a home security system. The accompanying comparison tool will document your research and provide ...

Software Purchase Approval Form

Software is a critical component of company operations, helping employees do their jobs so the business can meet its goals. This software purchase approval form, courtesy of TechRepublic Premium, can help you meet these commitments and get the most out of your applications. From the download: The form should be filled out via the following ...

• TechRepublic on Twitter
• TechRepublic on Facebook
• TechRepublic on LinkedIn
• TechRepublic on Flipboard
• Privacy Policy
• Terms of Use
• Property of TechnologyAdvice

This paper is in the following e-collection/theme issue:

Published on 31.8.2023 in Vol 25 (2023)

AI Is Changing the Landscape of Academic Writing: What Can Be Done? Authors’ Reply to: AI Increases the Pressure to Overhaul the Scientific Peer Review Process. Comment on “Artificial Intelligence Can Generate Fraudulent but Authentic-Looking Scientific Medical Articles: Pandora’s Box Has Been Opened”

Authors of this article:

Letter to the Editor

• Martin Májovský 1 , MD, PhD   ; 
• Tomas Mikolov 2 , MSc, PhD   ; 
• David Netuka 1 , MD, Prof Dr  

1 Department of Neurosurgery and Neurooncology, First Faculty of Medicine, Charles University, Prague, Czech Republic

2 Czech Institute of Informatics, Robotics and Cybernetics, Czech Technical University in Prague, Prague, Czech Republic

Corresponding Author:

Martin Májovský, MD, PhD

Department of Neurosurgery and Neurooncology

First Faculty of Medicine

Charles University

U Vojenské nemocnice 1200

Prague, 16000

Czech Republic

Phone: 420 973202963

Email: [email protected]

With every new technology, there are benefits that come hand in hand with the potential for misuse. Currently, many researchers are already using large language models (LLMs) for tasks such as language editing and abstract generation, which is beneficial and time-saving. It is undeniable that in the near future, we will witness dedicated tools that facilitate the preparation of scientific manuscripts, including tasks like reviewing relevant literature and generating discussion sections. However, an obvious danger arises: the possibility of fabricating entire articles using LLMs without conducting any genuine research [ 1 ]. It is important to note that LLMs do not enable scientists to cheat; scientific fraud has existed long before their advent, but LLMs simply make it easier.

As mentioned by Liu and Brown [ 2 ], the scientific community is under increasing pressure to overhaul the peer review and publishing processes. Solving complex problems is never a simple task. Should the use of LLMs be banned, or should we focus on detecting artificial intelligence (AI)–generated text? Banning the use of LLMs is a naive approach. Any AI-based tool designed to detect AI-generated text will inevitably fail because one can always train the next language model on the outputs of such a tool. So, what can be done? We believe that implementing the following measures, some of which are already in use, may help reduce the number of fraudulent papers generated by AI:

• Provision of source data sets publicly . Authors should provide anonymized data sets with all subject information with the submission as a supplementary material. This measure can discourage researchers from submitting completely fraudulent papers.
• A meticulous review process . Respectable publishers should emphasize a quality review process by selecting and educating high-profile reviewers. Some kind of reviewer award system may boost the motivation of potential reviewers. In addition to the award system, a reviewer-ranking system could be implemented based on the quality of the review reports.
• Strict ethical regulations at the level of publishers . Questions related to the use of LLMs should be incorporated into the publishing ethics questionnaire.
• Strict ethical regulations at the level of academic institutions . Researchers should be motivated at the level of their institutions (eg, universities) to publish high-quality, genuine research in esteemed journals.
• Penalties for researchers who commit ethical misconduct . Researchers who are convicted of ethical fraud should be penalized. This could be carried out by a temporary or permanent ban from publishing with certain publishers, limiting indexing, etc.

LLMs are dramatically transforming the landscape of academic writing. The scientific community should not resist the modern advances of generative AI but rather seek to accommodate them. Undoubtedly, this process will be tedious and challenging. We are pleased that our work [ 1 ] contributes to the important debate taking place across the scientific community regarding this topic.

Conflicts of Interest

None declared.

• Májovský M, Černý M, Kasal M, Komarc M, Netuka D. Artificial intelligence can generate fraudulent but authentic-looking scientific medical articles: Pandora’s box has been opened. J Med Internet Res 2023 May 31;25:e46924 [ https://www.jmir.org/2023//e46924/ ] [ CrossRef ] [ Medline ]
• Liu N, Brown A. AI Increases the Pressure to Overhaul the Scientific Peer Review Process. Comment on “Artificial Intelligence Can Generate Fraudulent but Authentic-Looking Scientific Medical Articles: Pandora’s Box Has Been Opened”. J Med Internet Res 2023;25:e50591 [ https://www.jmir.org/2023/1/e50591 ] [ CrossRef ]

Abbreviations

Edited by T Leung; This is a non–peer-reviewed article. submitted 14.07.23; accepted 12.08.23; published 31.08.23

©Martin Májovský, Tomas Mikolov, David Netuka. Originally published in the Journal of Medical Internet Research (https://www.jmir.org), 31.08.2023.

This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on https://www.jmir.org/, as well as this copyright and license information must be included.

• SUGGESTED TOPICS
• The Magazine
• Newsletters
• Managing Yourself
• Managing Teams
• Work-life Balance
• The Big Idea
• Data & Visuals
• Reading Lists
• Case Selections
• HBR Learning
• Topic Feeds
• Account Settings
• Email Preferences

Using ChatGPT to Make Better Decisions

• Thomas Ramge
• Viktor Mayer-Schönberger

While it may be tempting to just ask for answers, LLMs can assist you at every stage of the decision-making process.

A successful decision-making process has three steps: Framing the decision, generating alternatives, and deciding between them. Large language models can help at each stage of the process. But while it may be tempting to merely ask ChatGPT for answers, the real power of LLMs is how they can assist at each stage. Ask for help thinking of considerations you might be missing, or alternatives you might not have considered. LLMs can be a de-biasing tool, helping you frame and make the decision yourself.

Can ChatGPT help executives make better decisions? The large language model everyone has been talking about for months also has an eloquent answer to this question: “Yes, I can support you in management decisions by providing information, facts, analysis, and perspectives that can help you make an informed decision.” ChatGPT immediately follows up with a limitation of its own competence. “However, it is important to note that my advice and recommendations are based on an algorithmic analysis of data and information, and you, as a human being, still have to make the final decision based on your experience, knowledge, and assessment of the situation.”

Fair enough. But despite this dose of modesty — or because of it — large language models like ChatGPT can become powerful decision-making tools for managers and for companies. Their promise isn’t in providing us answers, but in helping us go through a more systematic decision-making process than is often the case today, even with important management decisions.

Three phases characterize well-informed decisions. First, we must define our goals and context. What exactly is the decision about, and based on which goals, values, and preferences? This way, we define the decision-making problem and set the decision-making framework. The second step is to develop choices: What decision-making options are available to us? The goal here is to generate many different alternatives and not, as is all too often the case, to focus just on the obvious options. Only when we have developed sufficient options from the decision-making framework can we evaluate them and make a well-informed decision in a third step.

Used skillfully, ChatGPT can already provide valuable services in all three phases for business decisions in its current training state. In practice, this means we can enter into a dialogue with the system on any of the three phases of a well-informed decision-making system. When evaluating decision-making alternatives, we can ask, for example: What mistakes do managing directors of large, medium-sized companies in mechanical engineering make when they decide to expand into new markets? And what were the success criteria for a successful expansion?

ChatGPT then does not provide us with a template with which we can weigh the options perfectly in our case. But it can help us uncover our own biases and challenge preconceived notions. Using ChatGPT cleverly can be like a de-biasing tool that has seemingly read Daniel Kahneman and Amos Tversky intensively. It thus offers food for thought to better reflect on how we can evaluate the options in a more well-informed way.

The system is already even more valuable today when it is employed to work out additional options that we can not think of or easily come up with. This way, it broadens our decision-making horizons, and we understand that there are many more and more far-reaching decision-making options than we realize.

How do we reduce our dependence on China and diversify a supply chain? A managing director and his team may never have dealt with this decision-making question before. ChatGPT, however, may be able to offer up many of the strategies documented on the internet by companies in a comparable situation and may come up with more original ideas than simply relocating production to Vietnam. This is because the system has access to a part of the publicly available treasure trove of options in the industry or company class.

Large language models can also help set goals and preferences, evaluate the decision-making circumstances, and select the decision-making framework. Again, dialogue is key. With the right questions, we become the interlocutor to better understand the context of a decision. For example, with ChatGPT, we can quickly see suggestions of what typical goals other companies might have had in mind in a comparable decision-making situation. For example, a prompt might look like this: “Hi ChatGPT, I am the head of a successful, mid-sized tooling manufacturer outside Columbus, Ohio. I am having difficulties attracting new talent, especially engineers. What may be the reasons for this? What strategies are similar manufacturing companies employing to cope with the talent shortage?”

The bottom line is: ChatGPT is becoming an increasingly intelligent conversation and sparring partner. It does not relieve us of defining the decision-making framework, working out a wide range of options, and evaluating them. However — and here, the self-assessment from the beginning of this article is correct — it does provide interesting perspectives. A large language model has several advantages compared to a human sparring partner: It does not pursue its own interests and does not want to please the top decision-maker, for example, to promote its own career. It is not subject to internal group thinking and bureaucratic politics and is also much cheaper than external management consultants or internal strategy departments . This also means that ChatGPT may make the preparation and assistance of decisions for smaller companies cheaper, leveling the playing field.

The future of case studies

Budding managers at business schools are already indirectly learning about decision-making through a large number of case studies. The aim is to acquire a repertoire of decision-making models by developing and evaluating possible options for action within a decision-making framework. Of course, case studies do not contain a solution in the form of a perfect answer to a specific decision-making situation. In case studies, questions are raised, decision-making frameworks are presented, and decision-making options are outlined. Not only can prospective managers learn from and with these case studies, but they can also be used to train large language models. However, this has not yet happened.

ChatGPT’s programmers could only feed their model a fraction of publicly available case studies. The real treasure trove of data is exclusive and stored at the major providers such as Harvard Business Publishing (HBR’s parent company), with over 50,000 case studies or the non-profit Case Center . If the custodians of these business case studies team up with the makers of large language models, a language assistant for programming, copywriting, and customer inquiries could turn into a powerful decision-making assistant for companies.

This will also get easier in the future because the learning algorithms are becoming more and more efficient, and thus “medium-sized language models” will also be possible, in which it is no longer necessary to feed half the Internet and entire libraries, but above all the texts and documents relevant to the specific field. It is only a matter of time before this happens. In any case, the economic incentive for more informed business decisions is excellent and will propel the transition from today’s ChatGPT to an even more powerful future we might dub “DecisionGPT.”

The great strength of ChatGPT and similar systems is to compare and contrast similar situations. This is precisely the most important need in many management decisions. Very few of the decisions managers face are unique. Thousands, sometimes even millions, of managers before them have had to make a similar choice. The better it is described in human language how they set the decision-making framework, weigh the options, and make their decision, the easier it is for DecisionGPT to become a powerful tool for more informed decision-making.

Eventually, many such management decisions could be automated. Robo-managers could be deployed sooner and more often than many executives in their corner offices may believe today.

In the meantime, though, the advantage will go to managers who use currently available tools to improve their decision-making process. Don’t ask models like ChatGPT for answers; probe them to each stage of the decision-making process.

• TR Thomas Ramge has authored more than fifteen books about technology, innovation, and decision-making and has won numerous publishing awards.
• VM Viktor Mayer-Schönberger is Professor of Internet Governance and Regulation at the University of Oxford.

Partner Center

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

• View all journals
• My Account Login
• Explore content
• About the journal
• Publish with us
• Sign up for alerts
• Open Access
• Published: 30 August 2023

Champion-level drone racing using deep reinforcement learning

• Elia Kaufmann   ORCID: orcid.org/0000-0001-6094-5901 1 ,
• Leonard Bauersfeld 1 ,
• Antonio Loquercio   ORCID: orcid.org/0000-0002-8410-3933 1 ,
• Matthias Müller 2 ,
• Vladlen Koltun 3 &
• Davide Scaramuzza   ORCID: orcid.org/0000-0002-3831-6778 1  

Nature volume  620 ,  pages 982–987 ( 2023 ) Cite this article

55k Accesses

1249 Altmetric

Metrics details

• Aerospace engineering
• Computer science
• Electrical and electronic engineering
• Mechanical engineering

First-person view (FPV) drone racing is a televised sport in which professional competitors pilot high-speed aircraft through a 3D circuit. Each pilot sees the environment from the perspective of their drone by means of video streamed from an onboard camera. Reaching the level of professional pilots with an autonomous drone is challenging because the robot needs to fly at its physical limits while estimating its speed and location in the circuit exclusively from onboard sensors 1 . Here we introduce Swift, an autonomous system that can race physical vehicles at the level of the human world champions. The system combines deep reinforcement learning (RL) in simulation with data collected in the physical world. Swift competed against three human champions, including the world champions of two international leagues, in real-world head-to-head races. Swift won several races against each of the human champions and demonstrated the fastest recorded race time. This work represents a milestone for mobile robotics and machine intelligence 2 , which may inspire the deployment of hybrid learning-based solutions in other physical systems.

Deep RL 3 has enabled some recent advances in artificial intelligence. Policies trained with deep RL have outperformed humans in complex competitive games, including Atari 4 , 5 , 6 , Go 5 , 7 , 8 , 9 , chess 5 , 9 , StarCraft 10 , Dota 2 (ref.  11 ) and Gran Turismo 12 , 13 . These impressive demonstrations of the capabilities of machine intelligence have primarily been limited to simulation and board-game environments, which support policy search in an exact replica of the testing conditions. Overcoming this limitation and demonstrating champion-level performance in physical competitions is a long-standing problem in autonomous mobile robotics and artificial intelligence 14 , 15 , 16 .

FPV drone racing is a televised sport in which highly trained human pilots push aerial vehicles to their physical limits in high-speed agile manoeuvres (Fig. 1a ). The vehicles used in FPV racing are quadcopters, which are among the most agile machines ever built (Fig. 1b ). During a race, the vehicles exert forces that surpass their own weight by a factor of five or more, reaching speeds of more than 100 km h −1 and accelerations several times that of gravity, even in confined spaces. Each vehicle is remotely controlled by a human pilot who wears a headset showing a video stream from an onboard camera, creating an immersive ‘first-person-view’ experience (Fig. 1c ).

a , Swift (blue) races head-to-head against Alex Vanover, the 2019 Drone Racing League world champion (red). The track comprises seven square gates that must be passed in order in each lap. To win a race, a competitor has to complete three consecutive laps before its opponent. b , A close-up view of Swift, illuminated with blue LEDs, and a human-piloted drone, illuminated with red LEDs. The autonomous drones used in this work rely only on onboard sensory measurements, with no support from external infrastructure, such as motion-capture systems. c , From left to right: Thomas Bitmatta, Marvin Schaepper and Alex Vanover racing their drones through the track. Each pilot wears a headset that shows a video stream transmitted in real time from a camera aboard their aircraft. The headsets provide an immersive ‘first-person-view’ experience. c , Photo by Regina Sablotny.

Attempts to create autonomous systems that reach the performance of human pilots date back to the first autonomous drone racing competition in 2016 (ref.  17 ). A series of innovations followed, including the use of deep networks to identify the next gate location 18 , 19 , 20 , transfer of racing policies from simulation to reality 21 , 22 and accounting for uncertainty in perception 23 , 24 . The 2019 AlphaPilot autonomous drone racing competition showcased some of the best research in the field 25 . However, the first two teams still took almost twice as long as a professional human pilot to complete the track 26 , 27 . More recently, autonomous systems have begun to reach expert human performance 28 , 29 , 30 . However, these works rely on near-perfect state estimation provided by an external motion-capture system. This makes the comparison with human pilots unfair, as humans only have access to onboard observations from the drone.

In this article, we describe Swift, an autonomous system that can race a quadrotor at the level of human world champions using only onboard sensors and computation. Swift consists of two key modules: (1) a perception system that translates high-dimensional visual and inertial information into a low-dimensional representation and (2) a control policy that ingests the low-dimensional representation produced by the perception system and produces control commands.

The control policy is represented by a feedforward neural network and is trained in simulation using model-free on-policy deep RL 31 . To bridge discrepancies in sensing and dynamics between simulation and the physical world, we make use of non-parametric empirical noise models estimated from data collected on the physical system. These empirical noise models have proved to be instrumental for successful transfer of the control policy from simulation to reality.

We evaluate Swift on a physical track designed by a professional drone-racing pilot (Fig. 1a ). The track comprises seven square gates arranged in a volume of 30 × 30 × 8 m, forming a lap of 75 m in length. Swift raced this track against three human champions: Alex Vanover, the 2019 Drone Racing League world champion, Thomas Bitmatta, two-time MultiGP International Open World Cup champion, and Marvin Schaepper, three-time Swiss national champion. The quadrotors used by Swift and by the human pilots have the same weight, shape and propulsion. They are similar to drones used in international competitions.

The human pilots were given one week of practice on the race track. After this week of practice, each pilot competed against Swift in several head-to-head races (Fig. 1a,b ). In each head-to-head race, two drones (one controlled by a human pilot and one controlled by Swift) start from a podium. The race is set off by an acoustic signal. The first vehicle that completes three full laps through the track, passing all gates in the correct order in each lap, wins the race.

Swift won several races against each of the human pilots and achieved the fastest race time recorded during the events. Our work marks the first time, to our knowledge, that an autonomous mobile robot achieved world-champion-level performance in a real-world competitive sport.

The Swift system

Swift uses a combination of learning-based and traditional algorithms to map onboard sensory readings to control commands. This mapping comprises two parts: (1) an observation policy, which distils high-dimensional visual and inertial information into a task-specific low-dimensional encoding, and (2) a control policy that transforms the encoding into commands for the drone. A schematic overview of the system is shown in Fig. 2 .

Swift consists of two key modules: a perception system that translates visual and inertial information into a low-dimensional state observation and a control policy that maps this state observation to control commands. Control commands specify desired collective thrust and body rates, the same control modality that the human pilots use. a , The perception system consists of a VIO module that computes a metric estimate of the drone state from camera images and high-frequency measurements obtained by an inertial measurement unit (IMU). The VIO estimate is coupled with a neural network that detects the corners of racing gates in the image stream. The corner detections are mapped to a 3D pose and fused with the VIO estimate using a Kalman filter. b , We use model-free on-policy deep RL to train the control policy in simulation. During training, the policy maximizes a reward that combines progress towards the centre of the next racing gate with a perception objective to keep the next gate in the field of view of the camera. To transfer the racing policy from simulation to the physical world, we augment the simulation with data-driven residual models of the vehicle’s perception and dynamics. These residual models are identified from real-world experience collected on the race track. MLP, multilayer perceptron.

The observation policy consists of a visual–inertial estimator 32 , 33 that operates together with a gate detector 26 , which is a convolutional neural network that detects the racing gates in the onboard images. Detected gates are then used to estimate the global position and orientation of the drone along the race track. This is done by a camera-resectioning algorithm 34 in combination with a map of the track. The estimate of the global pose obtained from the gate detector is then combined with the estimate from the visual–inertial estimator by means of a Kalman filter, resulting in a more accurate representation of the robot’s state. The control policy, represented by a two-layer perceptron, maps the output of the Kalman filter to control commands for the aircraft. The policy is trained using on-policy model-free deep RL 31 in simulation. During training, the policy maximizes a reward that combines progress towards the next racing gate 35 with a perception objective that rewards keeping the next gate in the field of view of the camera. Seeing the next gate is rewarded because it increases the accuracy of pose estimation.

Optimizing a policy purely in simulation yields poor performance on physical hardware if the discrepancies between simulation and reality are not mitigated. The discrepancies are caused primarily by two factors: (1) the difference between simulated and real dynamics and (2) the noisy estimation of the robot’s state by the observation policy when provided with real sensory data. We mitigate these discrepancies by collecting a small amount of data in the real world and using this data to increase the realism of the simulator.

Specifically, we record onboard sensory observations from the robot together with highly accurate pose estimates from a motion-capture system while the drone is racing through the track. During this data-collection phase, the robot is controlled by a policy trained in simulation that operates on the pose estimates provided by the motion-capture system. The recorded data allow to identify the characteristic failure modes of perception and dynamics observed through the race track. These intricacies of failing perception and unmodelled dynamics are dependent on the environment, platform, track and sensors. The perception and dynamics residuals are modelled using Gaussian processes 36 and k -nearest-neighbour regression, respectively. The motivation behind this choice is that we empirically found perception residuals to be stochastic and dynamics residuals to be largely deterministic (Extended Data Fig. 1 ). These residual models are integrated into the simulation and the racing policy is fine-tuned in this augmented simulation. This approach is related to the empirical actuator models used for simulation-to-reality transfer in ref.  37 but further incorporates empirical modelling of the perception system and also accounts for the stochasticity in the estimate of the platform state.

We ablate each component of Swift in controlled experiments reported in the extended data. Also, we compare against recent work that tackles the task of autonomous drone racing with traditional methods, including trajectory planning and model predictive control (MPC). Although such approaches achieve comparable or even superior performance to our approach in idealized conditions, such as simplified dynamics and perfect knowledge of the robot’s state, their performance collapses when their assumptions are violated. We find that approaches that rely on precomputed paths 28 , 29 are particularly sensitive to noisy perception and dynamics. No traditional method has achieved competitive lap times compared with Swift or human world champions, even when provided with highly accurate state estimation from a motion-capture system. Detailed analysis is provided in the extended data.

The drone races take place on a track designed by an external world-class FPV pilot. The track features characteristic and challenging manoeuvres, such as a Split-S (Figs. 1a (top-right corner) and 4d ). Pilots are allowed to continue racing even after a crash, provided their vehicle is still able to fly. If both drones crash and cannot complete the track, the drone that proceeded farther along the track wins.

As shown in Fig. 3b , Swift wins 5 out of 9 races against A. Vanover, 4 out of 7 races against T. Bitmatta and 6 out of 9 races against M. Schaepper. Out of the 10 losses recorded for Swift, 40% were because of a collision with the opponent, 40% because of collision with a gate and 20% because of the drone being slower than the human pilot. Overall, Swift wins most races against each human pilot. Swift also achieves the fastest race time recorded, with a lead of half a second over the best time clocked by a human pilot (A. Vanover).

a , Lap-time results. We compare Swift against the human pilots in time-trial races. Lap times indicate best single lap times and best average times achieved in a heat of three consecutive laps. The reported statistics are computed over a dataset recorded during one week on the race track, which corresponds to 483 (115) data points for Swift, 331 (221) for A. Vanover, 469 (338) for T. Bitmatta and 345 (202) for M. Schaepper. The first number is the number of single laps and the second is the number of three consecutive laps. The dark points in each distribution correspond to laps flown in race conditions. b , Head-to-head results. We report the number of head-to-head races flown by each pilot, the number of wins and losses, as well as the win ratio.

Figure 4 and Extended Data Table 1d provide an analysis of the fastest lap flown by Swift and each human pilot. Although Swift is globally faster than all human pilots, it is not faster on all individual segments of the track (Extended Data Table 1 ). Swift is consistently faster at the start and in tight turns such as the split S. At the start, Swift has a lower reaction time, taking off from the podium, on average, 120 ms before human pilots. Also, it accelerates faster and reaches higher speeds going into the first gate (Extended Data Table 1d , segment 1). In sharp turns, as shown in Fig. 4c,d , Swift finds tighter manoeuvres. One hypothesis is that Swift optimizes trajectories on a longer timescale than human pilots. It is known that model-free RL can optimize long-term rewards through a value function 38 . Conversely, human pilots plan their motion on a shorter timescale, up to one gate into the future 39 . This is apparent, for example in the split S (Fig. 4b,d ), for which human pilots are faster in the beginning and at the end of the manoeuvre, but slower overall (Extended Data Table 1d , segment 3). Also, human pilots orient the aircraft to face the next gate earlier than Swift does (Fig. 4c,d ). We propose that human pilots are accustomed to keeping the upcoming gate in view, whereas Swift has learned to execute some manoeuvres while relying on other cues, such as inertial data and visual odometry against features in the surrounding environments. Overall, averaged over the entire track, the autonomous drone achieves the highest average speed, finds the shortest racing line and manages to maintain the aircraft closer to its actuation limits throughout the race, as indicated by the average thrust and power drawn (Extended Data Table 1d ).

a , Comparison of the fastest race of each pilot, illustrated by the time behind Swift. The time difference from the autonomous drone is computed as the time since it passed the same position on the track. Although Swift is globally faster than all human pilots, it is not necessarily faster on all individual segments of the track. b , Visualization of where the human pilots are faster (red) and slower (blue) compared with the autonomous drone. Swift is consistently faster at the start and in tight turns, such as the split S. c , Analysis of the manoeuvre after gate 2. Swift in blue, Vanover in red. Swift gains time against human pilots in this segment as it executes a tighter turn while maintaining comparable speed. d , Analysis of the split S manoeuvre. Swift in blue, Vanover in red. The split S is the most challenging segment in the race track, requiring a carefully coordinated roll and pitch motion that yields a descending half-loop through the two gates. Swift gains time against human pilots on this segment as it executes a tighter turn with less overshoot. e , Illustration of track segments used for analysis. Segment 1 is traversed once at the start, whereas segments 2–4 are traversed in each lap (three times over the course of a race).

We also compare the performance of Swift and the human champions in time trials (Fig. 3a ). In a time trial, a single pilot races the track, with the number of laps left to the discretion of the pilot. We accumulate time-trial data from the practice week and the races, including training runs (Fig. 3a , coloured) and laps flown in race conditions (Fig. 3a , black). For each contestant, we use more than 300 laps for computing statistics. The autonomous drone more consistently pushes for fast lap times, exhibiting lower mean and variance. Conversely, human pilots decide whether to push for speed on a lap-by-lap basis, yielding higher mean and variance in lap times, both during training and in the races. The ability to adapt the flight strategy allows human pilots to maintain a slower pace if they identify that they have a clear lead, so as to reduce the risk of a crash. The autonomous drone is unaware of its opponent and pushes for fastest expected completion time no matter what, potentially risking too much when in the lead and too little when trailing behind 40 .

FPV drone racing requires real-time decision-making based on noisy and incomplete sensory input from the physical environment. We have presented an autonomous physical system that achieves champion-level performance in this sport, reaching—and at times exceeding—the performance of human world champions. Our system has certain structural advantages over the human pilots. First, it makes use of inertial data from an onboard inertial measurement unit 32 . This is akin to the human vestibular system 41 , which is not used by the human pilots because they are not physically in the aircraft and do not feel the accelerations acting on it. Second, our system benefits from lower sensorimotor latency (40 ms for Swift versus an average of 220 ms for expert human pilots 39 ). On the other hand, the limited refresh rate of the camera used by Swift (30 Hz) can be considered a structural advantage for human pilots, whose cameras’ refresh rate is four times as fast (120 Hz), improving their reaction time 42 .

Human pilots are impressively robust: they can crash at full speed, and—if the hardware still functions—carry on flying and complete the track. Swift was not trained to recover after a crash. Human pilots are also robust to changes in environmental conditions, such as illumination, which can markedly alter the appearance of the track. By contrast, Swift’s perception system assumes that the appearance of the environment is consistent with what was observed during training. If this assumption fails, the system can fail. Robustness to appearance changes can be provided by training the gate detector and the residual observation model in a diverse set of conditions. Addressing these limitations could enable applying the presented approach in autonomous drone racing competitions in which access to the environment and the drone is limited 25 .

Notwithstanding the remaining limitations and the work ahead, the attainment by an autonomous mobile robot of world-champion-level performance in a popular physical sport is a milestone for robotics and machine intelligence. This work may inspire the deployment of hybrid learning-based solutions in other physical systems, such as autonomous ground vehicles, aircraft and personal robots, across a broad range of applications.

Quadrotor simulation

Quadrotor dynamics.

To enable large-scale training, we use a high-fidelity simulation of the quadrotor dynamics. This section briefly explains the simulation. The dynamics of the vehicle can be written as

in which ⊙ represents quaternion rotation, \({{\bf{p}}}_{{\mathcal{W}}{\mathcal{B}}},{{\bf{q}}}_{{\mathcal{W}}{\mathcal{B}}},{{\bf{v}}}_{{\mathcal{W}}}\) and \({{\boldsymbol{\omega }}}_{{\mathcal{B}}}\) denote the position, attitude quaternion, inertial velocity and body rates of the quadcopter, respectively. The motor time constant is k mot and the motor speeds Ω and Ω ss are the actual and steady-state motor speeds, respectively. The matrix J  is the inertia of the quadcopter and \({{\bf{g}}}_{{\mathcal{W}}}\) denotes the gravity vector. Two forces act on the quadrotor: the lift force f prop generated by the propellers and an aerodynamic force f aero that aggregates all other forces, such as aerodynamic drag, dynamic lift and induced drag. The torque is modelled as a sum of four components: the torque generated by the individual propeller thrusts τ prop , the yaw torque τ mot generated by a change in motor speed, an aerodynamic torque τ aero that accounts for various aerodynamic effects such as blade flapping and an inertial term τ iner . The individual components are given as

in which r P, i is the location of propeller i , expressed in the body frame, and f i and τ i are the forces and torques, respectively, generated by the i th propeller. The axis of rotation of the i th motor is denoted by ζ i , the combined inertia of the motor and propeller is J m+p and the derivative of the i th motor speed is \({\dot{\Omega }}_{i}\) . The individual propellers are modelled using a commonly used quadratic model, which assumes that the lift force and drag torque are proportional to the square of the propeller speed Ω i :

in which c l and c d denote the propeller lift and drag coefficients, respectively.

Aerodynamic forces and torques

The aerodynamic forces and torques are difficult to model with a first-principles approach. We thus use a data-driven model 43 . To maintain the low computational complexity required for large-scale RL training, a grey-box polynomial model is used rather than a neural network. The aerodynamic effects are assumed to primarily depend on the velocity \({{\bf{v}}}_{{\mathcal{B}}}\) (in the body frame) and the average squared motor speed \(\overline{{\Omega }^{2}}\) . The aerodynamic forces f x , f y and f z and torques τ x , τ y and τ z are estimated in the body frame. The quantities v x , v y and v z denote the three axial velocity components (in the body frame) and v x y denotes the speed in the ( x ,  y ) plane of the quadrotor. On the basis of insights from the underlying physical processes, linear and quadratic combinations of the individual terms are selected. For readability, the coefficients multiplying each summand have been omitted:

The respective coefficients are then identified from real-world flight data, in which motion capture is used to provide ground-truth forces and torque measurements. We use data from the race track, allowing the dynamics model to fit the track. This is akin to the human pilots’ training for days or weeks before the race on the specific track that they will be racing. In our case, the human pilots are given a week of practice on the same track ahead of the competition.

Betaflight low-level controller

To control the quadrotor, the neural network outputs collective thrust and body rates. This control signal is known to combine high agility with good robustness to simulation-to-reality transfer 44 . The predicted collective thrust and body rates are then processed by an onboard low-level controller that computes individual motor commands, which are subsequently translated into analogue voltage signals through an electronic speed controller (ESC) that controls the motors. On the physical vehicle, this low-level proportional–integral–derivative (PID) controller and ESC are implemented using the open-source Betaflight and BLHeli32 firmware 45 . In simulation, we use an accurate model of both the low-level controller and the motor speed controller.

Because the Betaflight PID controller has been optimized for human-piloted flight, it exhibits some peculiarities, which the simulation correctly captures: the reference for the D-term is constantly zero (pure damping), the I-term gets reset when the throttle is cut and, under motor thrust saturation, the body rate control is assigned priority (proportional downscaling of all motor signals to avoid saturation). The gains of the controller used for simulation have been identified from the detailed logs of the Betaflight controller’s internal states. The simulation can predict the individual motor commands with less than 1% error.

Battery model and ESC

The low-level controller converts the individual motor commands into a pulse-width modulation (PWM) signal and sends it to the ESC, which controls the motors. Because the ESC does not perform closed-loop control of the motor speeds, the steady-state motor speed Ω i ,ss for a given PWM motor command cmd i is a function of the battery voltage. Our simulation thus models the battery voltage using a grey-box battery model 46 that simulates the voltage based on instantaneous power consumption P mot :

The battery model 46 then simulates the battery voltage based on this power demand. Given the battery voltage U bat and the individual motor command u cmd, i , we use the mapping (again omitting the coefficients multiplying each summand)

to calculate the corresponding steady-state motor speed Ω i ,ss required for the dynamics simulation in equation ( 1 ). The coefficients have been identified from Betaflight logs containing measurements of all involved quantities. Together with the model of the low-level controller, this enables the simulator to correctly translate an action in the form of collective thrust and body rates to desired motor speeds Ω ss in equation ( 1 ).

Policy training

We train deep neural control policies that directly map observations o t in the form of platform state and next gate observation to control actions u t in the form of mass-normalized collective thrust and body rates 44 . The control policies are trained using model-free RL in simulation.

Training algorithm

Training is performed using proximal policy optimization 31 . This actor-critic approach requires jointly optimizing two neural networks during training: the policy network, which maps observations to actions, and the value network, which serves as the ‘critic’ and evaluates actions taken by the policy. After training, only the policy network is deployed on the robot.

Observations, actions and rewards

An observation \({{\bf{o}}}_{t}\in {{\mathbb{R}}}^{31}\) obtained from the environment at time t consists of: (1) an estimate of the current robot state; (2) the relative pose of the next gate to be passed on the track layout; and (3) the action applied in the previous step. Specifically, the estimate of the robot state contains the position of the platform, its velocity and attitude represented by a rotation matrix, resulting in a vector in \({{\mathbb{R}}}^{15}\) . Although the simulation uses quaternions internally, we use a rotation matrix to represent attitude to avoid ambiguities 47 . The relative pose of the next gate is encoded by providing the relative position of the four gate corners with respect to the vehicle, resulting in a vector in \({{\mathbb{R}}}^{12}\) . All observations are normalized before being passed to the network. Because the value network is only used during training time, it can access privileged information about the environment that is not accessible to the policy 48 . This privileged information is concatenated with other inputs to the policy network and contains the exact position, orientation and velocity of the robot.

For each observation o t , the policy network produces an action \({{\bf{a}}}_{t}\in {{\mathbb{R}}}^{4}\) in the form of desired mass-normalized collective thrust and body rates.

We use a dense shaped reward formulation to learn the task of perception-aware autonomous drone racing. The reward r t at time step t is given by

in which r prog rewards progress towards the next gate 35 , r perc encodes perception awareness by adjusting the attitude of the vehicle such that the optical axis of the camera points towards the centre of the next gate, r cmd rewards smooth actions and r crash is a binary penalty that is only active when colliding with a gate or when the platform leaves a predefined bounding box. If r crash is triggered, the training episode ends.

Specifically, the reward terms are

in which \({d}_{t}^{{\rm{Gate}}}\) denotes the distance from the centre of mass of the vehicle to the centre of the next gate at time step t , δ cam represents the angle between the optical axis of the camera and the centre of the next gate and \({{\bf{a}}}_{t}^{\omega }\) are the commanded body rates. The hyperparameters λ 1 ,…,  λ 5 balance different terms (Extended Data Table 1a ).

Training details

Data collection is performed by simulating 100 agents in parallel that interact with the environment in episodes of 1,500 steps. At each environment reset, every agent is initialized at a random gate on the track, with bounded perturbation around a state previously observed when passing this gate. In contrast to previous work 44 , 49 , 50 , we do not perform randomization of the platform dynamics at training time. Instead, we perform fine-tuning based on real-world data. The training environment is implemented using TensorFlow Agents 51 . The policy network and the value network are both represented by two-layer perceptrons with 128 nodes in each layer and LeakyReLU activations with a negative slope of 0.2. Network parameters are optimized using the Adam optimizer with learning rate 3 × 10 −4 for both the policy network and the value network.

Policies are trained for a total of 1 × 10 8 environment interactions, which takes 50 min on a workstation (i9 12900K, RTX 3090, 32 GB RAM DDR5). Fine-tuning is performed for 2 × 10 7 environment interactions.

Residual model identification

We perform fine-tuning of the original policy based on a small amount of data collected in the real world. Specifically, we collect three full rollouts in the real world, corresponding to approximately 50 s of flight time. We fine-tune the policy by identifying residual observations and residual dynamics, which are then used for training in simulation. During this fine-tuning phase, only the weights of the control policy are updated, whereas the weights of the gate-detection network are kept constant.

Residual observation model

Navigating at high speeds results in substantial motion blur, which can lead to a loss of tracked visual features and severe drift in linear odometry estimates. We fine-tune policies with an odometry model that is identified from only a handful of trials recorded in the real world. To model the drift in odometry, we use Gaussian processes 36 , as they allow fitting a posterior distribution of odometry perturbations, from which we can sample temporally consistent realizations.

Specifically, the Gaussian process model fits residual position, velocity and attitude as a function of the ground-truth robot state. The observation residuals are identified by comparing the observed visual–inertial odometry (VIO) estimates during a real-world rollout with the ground-truth platform states, which are obtained from an external motion-tracking system.

We treat each dimension of the observation separately, effectively fitting a set of nine 1D Gaussian processes to the observation residuals. We use a mixture of radial basis function kernels

in which L  is the diagonal length scale matrix and σ f and σ n represent the data and prior noise variance, respectively, and z i and z j represent data features. The kernel hyperparameters are optimized by maximizing the log marginal likelihood. After kernel hyperparameter optimization, we sample new realizations from the posterior distribution that are then used during fine-tuning of the policy. Extended Data Fig. 1 illustrates the residual observations in position, velocity and attitude in real-world rollouts, as well as 100 sampled realizations from the Gaussian process model.

Residual dynamics model

We use a residual model to complement the simulated robot dynamics 52 . Specifically, we identify residual accelerations as a function of the platform state s and the commanded mass-normalized collective thrust c :

We use k -nearest neighbour regression with k  = 5. The size of the dataset used for residual dynamics model identification depends on the track layout and ranges between 800 and 1,000 samples for the track layout used in this work.

Gate detection

To correct for drift accumulated by the VIO pipeline, the gates are used as distinct landmarks for relative localization. Specifically, gates are detected in the onboard camera view by segmenting gate corners 26 . The greyscale images provided by the Intel RealSense Tracking Camera T265 are used as input images for the gate detector. The architecture of the segmentation network is a six-level U-Net 53 with (8, 16, 16, 16, 16, 16) convolutional filters of size (3, 3, 3, 5, 7, 7) per level and a final extra layer operating on the output of the U-Net containing 12 filters. As the activation function, LeakyReLU with α  = 0.01 is used. For deployment on the NVIDIA Jetson TX2, the network is ported to TensorRT. To optimize memory footprint and computation time, inference is performed in half-precision mode (FP16) and images are downsampled to size 384 × 384 before being fed to the network. One forward pass through the network takes 40 ms on the NVIDIA Jetson TX2.

VIO drift estimation

The odometry estimates from the VIO pipeline 54 exhibit substantial drift during high-speed flight. We use gate detection to stabilize the pose estimates produced by VIO. The gate detector outputs the coordinates of the corners of all visible gates. A relative pose is first estimated for all predicted gates using infinitesimal plane-based pose estimation (IPPE) 34 . Given this relative pose estimate, each gate observation is assigned to the closest gate in the known track layout, thus yielding a pose estimate for the drone.

Owing to the low frequency of the gate detections and the high quality of the VIO orientation estimate, we only refine the translational components of the VIO measurements. We estimate and correct for the drift of the VIO pipeline using a Kalman filter that estimates the translational drift p d (position offset) and its derivative, the drift velocity v d . The drift correction is performed by subtracting the estimated drift states p d and v d from the corresponding VIO estimates. The Kalman filter state x is given by \({\bf{x}}={[{{\bf{p}}}_{{\rm{d}}}^{\top },{{\bf{v}}}_{{\rm{d}}}^{\top }]}^{\top }\in {{\mathbb{R}}}^{6}\) .

The state x and covariance P  updates are given by:

On the basis of measurements, the process noise is set to σ pos  = 0.05 and σ vel  = 0.1. The filter state and covariance are initialized to zero. For each measurement z k (pose estimate from a gate detection), the predicted VIO drift \({{\bf{x}}}_{k}^{-}\) is corrected to the estimate \({{\bf{x}}}_{k}^{+}\) according to the Kalman filter equations:

in which K k  is the Kalman gain, R  is the measurement covariance and H k is the measurement matrix. If several gates have been detected in a single camera frame, all relative pose estimates are stacked and processed in the same Kalman filter update step. The main source of measurement error is the uncertainty in the gate-corner detection of the network. This error in the image plane results in a pose error when IPPE is applied. We opted for a sampling-based approach to estimate the pose error from the known average gate-corner-detection uncertainty. For each gate, the IPPE algorithm is applied to the nominal gate observation as well as to 20 perturbed gate-corner estimates. The resulting distribution of pose estimates is then used to approximate the measurement covariance R  of the gate observation.

Simulation results

Reaching champion-level performance in autonomous drone racing requires overcoming two challenges: imperfect perception and incomplete models of the system’s dynamics. In controlled experiments in simulation, we assess the robustness of our approach to both of these challenges. To this end, we evaluate performance in a racing task when deployed in four different settings. In setting (1), we simulate a simplistic quadrotor model with access to ground-truth state observations. In setting (2), we replace the ground-truth state observations with noisy observations identified from real-world flights. These noisy observations are generated by sampling one realization from the residual observation model and are independent of the perception awareness of the deployed controller. Settings (3) and (4) share the observation models with the previous two settings, respectively, but replace the simplistic dynamics model with more accurate aerodynamical simulation 43 . These four settings allow controlled assessment of the sensitivity of the approach to changes in the dynamics and the observation fidelity.

In all four settings, we benchmark our approach against the following baselines: zero-shot, domain randomization and time-optimal. The zero-shot baseline represents a learning-based racing policy 35 trained using model-free RL that is deployed zero-shot from the training domain to the test domain. The training domain of the policy is equal to experimental setting (1), that is, idealized dynamics and ground-truth observations. Domain randomization extends the learning strategy from the zero-shot baseline by randomizing observations and dynamics properties to increase robustness. The time-optimal baseline uses a precomputed time-optimal trajectory 28 that is tracked using an MPC controller. This approach has shown the best performance in comparison with other model-based methods for time-optimal flight 55 , 56 . The dynamics model used by the trajectory generation and the MPC controller matches the simulated dynamics of experimental setting (1).

Performance is assessed by evaluating the fastest lap time, the average and minimum observed gate margin of successfully passed gates and the percentage of track successfully completed. The gate margin metric measures the distance between the drone and the closest point on the gate when crossing the gate plane. A high gate margin indicates that the quadrotor passed close to the centre of the gate. Leaving a smaller gate margin can increase speed but can also increase the risk of collision or missing the gate. Any lap that results in a crash is not considered valid.

The results are summarized in Extended Data Table 1c . All approaches manage to successfully complete the task when deployed in idealized dynamics and ground-truth observations, with the time-optimal baseline yielding the lowest lap time. When deployed in settings that feature domain shift, either in the dynamics or the observations, the performance of all baselines collapses and none of the three baselines are able to complete even a single lap. This performance drop is exhibited by both learning-based and traditional approaches. By contrast, our approach, which features empirical models of dynamics and observation noise, succeeds in all deployment settings, with small increases in lap time.

The key feature that enables our approach to succeed across deployment regimes is the use of an empirical model of dynamics and observation noise, estimated from real-world data. A comparison between an approach that has access to such data and approaches that do not is not entirely fair. For that reason, we also benchmark the performance of all baseline approaches when having access to the same real-world data used by our approach. Specifically, we compare the performance in experimental setting (2), which features the idealized dynamics model but noisy perception. All baseline approaches are provided with the predictions of the same Gaussian process model that we use to characterize observation noise. The results are summarized in Extended Data Table 1b . All baselines benefit from the more realistic observations, yielding higher completion rates. Nevertheless, our approach is the only one that reliably completes the entire track. As well as the predictions of the observation noise model, our approach also takes into account the uncertainty of the model. For an in-depth comparison of the performance of RL versus optimal control in controlled experiments, we refer the reader to ref.  57 .

Fine-tuning for several iterations

We investigate the extent of variations in behaviour across iterations. The findings of our analysis reveal that subsequent fine-tuning operations result in negligible enhancements in performance and alterations in behaviour (Extended Data Fig. 2 ).

In the following, we provide more details on this investigation. We start by enumerating the fine-tuning steps to provide the necessary notation:

Train policy-0 in simulation.

Deploy policy-0 in the real world. The policy operates on ground-truth data from a motion-capture system.

Identify residuals observed by policy-0 in the real world.

Train policy-1 by fine-tuning policy-0 on the identified residuals.

Deploy policy-1 in the real world. The policy operates only on onboard sensory measurements.

Identify residuals observed by policy-1 in the real world.

Train policy-2 by fine-tuning policy-1 on the identified residuals.

We compare the performance of policy-1 and policy-2 in simulation after fine-tuning on their respective residuals. The results are illustrated in Extended Data Fig. 2 . We observe that the difference in distance from gate centres, which is a metric for the safety of the policy, is 0.09 ± 0.08 m. Furthermore, the difference in the time taken to complete a single lap is 0.02 ± 0.02 s. Note that this lap-time difference is substantially smaller than the difference between the single-lap completion times of Swift and the human pilots (0.16 s).

Drone hardware configuration

The quadrotors used by the human pilots and Swift have the same weight, shape and propulsion. The platform design is based on the Agilicious framework 58 . Each vehicle has a weight of 870 g and can produce a maximum static thrust of approximately 35 N, which results in a static thrust-to-weight ratio of 4.1. The base of each platform consists of an Armattan Chameleon 6″ main frame that is equipped with T-Motor Velox 2306 motors and 5″, three-bladed propellers. An NVIDIA Jetson TX2 accompanied by a Connect Tech Quasar carrier board provides the main compute resource for the autonomous drones, featuring a six-core CPU running at 2 GHz and a dedicated GPU with 256 CUDA cores running at 1.3 GHz. Although forward passes of the gate-detection network are performed on the GPU, the racing policy is evaluated on the CPU, with one inference pass taking 8 ms. The autonomous drones carry an Intel RealSense Tracking Camera T265 that provides VIO estimates 59 at 100 Hz that are fed by USB to the NVIDIA Jetson TX2. The human-piloted drones carry neither a Jetson computer nor a RealSense camera and are instead equipped with a corresponding ballast weight. Control commands in the form of collective thrust and body rates produced by the human pilots or Swift are sent to a commercial flight controller, which runs on an STM32 processor operating at 216 MHz. The flight controller is running Betaflight, an open-source flight-control software 45 .

Human pilot impressions

The following quotes convey the impressions of the three human champions who raced against Swift.

Alex Vanover :

These races will be decided at the split S, it is the most challenging part of the track.

This was the best race! I was so close to the autonomous drone, I could really feel the turbulence when trying to keep up with it.

Thomas Bitmatta :

The possibilities are endless, this is the start of something that could change the whole world. On the flip side, I’m a racer, I don’t want anything to be faster than me.

As you fly faster, you trade off precision for speed.

It’s inspiring to see the potential of what drones are actually capable of. Soon, the AI drone could even be used as a training tool to understand what would be possible.

Marvin Schaepper :

It feels different racing against a machine, because you know that the machine doesn’t get tired.

Research ethics

The study has been conducted in accordance with the Declaration of Helsinki. The study protocol is exempt from review by an ethics committee according to the rules and regulations of the University of Zurich, because no health-related data has been collected. The participants gave their written informed consent before participating in the study.

Data availability

All (other) data needed to evaluate the conclusions in the paper are present in the paper or the extended data. Motion-capture recordings of the race events with accompanying analysis code can be found in the file ‘racing_data.zip’ on Zenodo at https://doi.org/10.5281/zenodo.7955278 .

Code availability

Pseudocode for Swift detailing the training process and algorithms can be found in the file ‘pseudocode.zip’ on Zenodo at https://doi.org/10.5281/zenodo.7955278 . To safeguard against potential misuse, the full source code associated with this research will not be made publicly available.

De Wagter, C., Paredes-Vallés, F., Sheth, N. & de Croon, G. Learning fast in autonomous drone racing. Nat. Mach. Intell. 3 , 923 (2021).

Article   Google Scholar  

Hanover, D. et al. Autonomous drone racing: a survey. Preprint at https://arxiv.org/abs/2301.01755 (2023).

Sutton, R. S. & Barto, A. G. Reinforcement Learning: An Introduction (MIT Press, 2018).

Mnih, V. et al. Human-level control through deep reinforcement learning. Nature 518 , 529–533 (2015).

Article   ADS   CAS   PubMed   Google Scholar  

Schrittwieser, J. et al. Mastering Atari, Go, chess and shogi by planning with a learned model. Nature 588 , 604–609 (2020).

Ecoffet, A., Huizinga, J., Lehman, J., Stanley, K. O. & Clune, J. First return, then explore. Nature 590 , 580–586 (2021).

Silver, D. et al. Mastering the game of Go with deep neural networks and tree search. Nature 529 , 484–489 (2016).

Silver, D. et al. Mastering the game of Go without human knowledge. Nature 550 , 354–359 (2017).

Silver, D. et al. A general reinforcement learning algorithm that masters chess, shogi, and Go through self-play. Science 362 , 1140–1144 (2018).

Article   ADS   MathSciNet   CAS   PubMed   MATH   Google Scholar  

Vinyals, O. et al. Grandmaster level in StarCraft II using multi-agent reinforcement learning. Nature 575 , 350–354 (2019).

Berner, C. et al. Dota 2 with large scale deep reinforcement learning. Preprint at https://arxiv.org/abs/1912.06680 (2019).

Fuchs, F., Song, Y., Kaufmann, E., Scaramuzza, D. & Dürr, P. Super-human performance in Gran Turismo Sport using deep reinforcement learning. IEEE Robot. Autom. Lett. 6 , 4257–4264 (2021).

Wurman, P. R. et al. Outracing champion Gran Turismo drivers with deep reinforcement learning. Nature 602 , 223–228 (2022).

Funke, J. et al. in Proc. 2012 IEEE Intelligent Vehicles Symposium 541–547 (IEEE, 2012).

Spielberg, N. A., Brown, M., Kapania, N. R., Kegelman, J. C. & Gerdes, J. C. Neural network vehicle models for high-performance automated driving. Sci. Robot. 4 , eaaw1975 (2019).

Article   PubMed   Google Scholar  

Won, D.-O., Müller, K.-R. & Lee, S.-W. An adaptive deep reinforcement learning framework enables curling robots with human-like performance in real-world conditions. Sci. Robot. 5 , eabb9764 (2020).

Moon, H., Sun, Y., Baltes, J. & Kim, S. J. The IROS 2016 competitions. IEEE Robot. Autom. Mag. 24 , 20–29 (2017).

Jung, S., Hwang, S., Shin, H. & Shim, D. H. Perception, guidance, and navigation for indoor autonomous drone racing using deep learning. IEEE Robot. Autom. Lett. 3 , 2539–2544 (2018).

Kaufmann, E. et al. in Proc. 2nd Conference on Robot Learning (CoRL) 133–145 (PMLR, 2018).

Zhang, D. & Doyle, D. D. in Proc. 2020 IEEE Aerospace Conference , 1–11 (IEEE, 2020).

Loquercio, A. et al. Deep drone racing: from simulation to reality with domain randomization. IEEE Trans. Robot. 36 , 1–14 (2019).

Loquercio, A. et al. Learning high-speed flight in the wild. Sci. Robot. 6 , eabg5810 (2021).

Kaufmann, E. et al. in Proc. 2019 International Conference on Robotics and Automation (ICRA) 690–696 (IEEE, 2019).

Li, S., van der Horst, E., Duernay, P., De Wagter, C. & de Croon, G. C. Visual model-predictive localization for computationally efficient autonomous racing of a 72-g drone. J. Field Robot. 37 , 667–692 (2020).

A.I. is flying drones (very, very slowly). https://www.nytimes.com/2019/03/26/technology/alphapilot-ai-drone-racing.html (2019).

Foehn, P. et al. AlphaPilot: autonomous drone racing. Auton. Robots 46 , 307–320 (2021).

Article   PubMed   PubMed Central   Google Scholar  

Wagter, C. D., Paredes-Vallé, F., Sheth, N. & de Croon, G. The sensing, state-estimation, and control behind the winning entry to the 2019 Artificial Intelligence Robotic Racing Competition. Field Robot. 2 , 1263–1290 (2022).

Foehn, P., Romero, A. & Scaramuzza, D. Time-optimal planning for quadrotor waypoint flight. Sci. Robot. 6 , eabh1221 (2021).

Romero, A., Sun, S., Foehn, P. & Scaramuzza, D. Model predictive contouring control for time-optimal quadrotor flight. IEEE Trans. Robot. 38 , 3340–3356 (2022).

Sun, S., Romero, A., Foehn, P., Kaufmann, E. & Scaramuzza, D. A comparative study of nonlinear MPC and differential-flatness-based control for quadrotor agile flight. IEEE Trans. Robot. 38 , 3357–3373 (2021).

Schulman, J., Wolski, F., Dhariwal, P., Radford, A. & Klimov, O. Proximal policy optimization algorithms. Preprint at https://arxiv.org/abs/1707.06347 (2017).

Scaramuzza, D. & Zhang, Z. Encyclopedia of Robotics (eds Ang, M., Khatib, O. & Siciliano, B.) 1–9 (Springer, 2019).

Huang, G. in Proc. 2019 International Conference on Robotics and Automation (ICRA) 9572–9582 (IEEE, 2019).

Collins, T. & Bartoli, A. Infinitesimal plane-based pose estimation. Int. J. Comput. Vis. 109 , 252–286 (2014).

Article   MathSciNet   MATH   Google Scholar  

Song, Y., Steinweg, M., Kaufmann, E. & Scaramuzza, D. in Proc. 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 1205–1212 (IEEE, 2021).

Williams, C. K. & Rasmussen, C. E. Gaussian Processes for Machine Learning (MIT Press, 2006).

Hwangbo, J. et al. Learning agile and dynamic motor skills for legged robots. Sci. Robot. 4 , eaau5872 (2019).

Hung, C.-C. et al. Optimizing agent behavior over long time scales by transporting value. Nat. Commun. 10 , 5223 (2019).

Article   ADS   PubMed   PubMed Central   Google Scholar  

Pfeiffer, C. & Scaramuzza, D. Human-piloted drone racing: visual processing and control. IEEE Robot. Autom. Lett. 6 , 3467–3474 (2021).

Spica, R., Cristofalo, E., Wang, Z., Montijano, E. & Schwager, M. A real-time game theoretic planner for autonomous two-player drone racing. IEEE Trans. Robot. 36 , 1389–1403 (2020).

Day, B. L. & Fitzpatrick, R. C. The vestibular system. Curr. Biol. 15 , R583–R586 (2005).

Article   CAS   PubMed   Google Scholar  

Kim, J. et al. Esports arms race: latency and refresh rate for competitive gaming tasks. J. Vis. 19 , 218c (2019).

Bauersfeld, L., Kaufmann, E., Foehn, P., Sun, S. & Scaramuzza, D. in Proc. Robotics: Science and Systems XVII 42 (Robotics: Science and Systems Foundation, 2021).

Kaufmann, E., Bauersfeld, L. & Scaramuzza, D. in Proc. 2022 International Conference on Robotics and Automation (ICRA) 10504–10510 (IEEE, 2022).

The Betaflight Open Source Flight Controller Firmware Project. Betaflight. https://github.com/betaflight/betaflight (2022).

Bauersfeld, L. & Scaramuzza, D. Range, endurance, and optimal speed estimates for multicopters. IEEE Robot. Autom. Lett. 7 , 2953–2960 (2022).

Zhou, Y., Barnes, C., Lu, J., Yang, J. & Li, H. in Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 5745–5753 (IEEE, 2019).

Pinto, L., Andrychowicz, M., Welinder, P., Zaremba, W. & Abbeel, P. in Proc. Robotics: Science and Systems XIV (MIT Press Journals, 2018).

Molchanov, A. et al. in Proc. 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 59–66 (IEEE, 2019).

Andrychowicz, O. M. et al. Learning dexterous in-hand manipulation. Int. J. Robot. Res. 39 , 3–20 (2020).

Guadarrama, S. et al. TF-Agents: a library for reinforcement learning in TensorFlow. https://github.com/tensorflow/agents (2018).

Torrente, G., Kaufmann, E., Foehn, P. & Scaramuzza, D. Data-driven MPC for quadrotors. IEEE Robot. Autom. Lett. 6 , 3769–3776 (2021).

Ronneberger, O., Fischer, P. & Brox, T. in Proc. International Conference on Medical Image Computing and Computer-assisted Intervention 234–241 (Springer, 2015).

Intel RealSense T265 series product family. https://www.intelrealsense.com/wp-content/uploads/2019/09/Intel_RealSense_Tracking_Camera_Datasheet_Rev004_release.pdf (2019).

Ryou, G., Tal, E. & Karaman, S. Multi-fidelity black-box optimization for time-optimal quadrotor maneuvers. Int. J. Robot. Res. 40 , 1352–1369 (2021).

Pham, H. & Pham, Q.-C. A new approach to time-optimal path parameterization based on reachability analysis. IEEE Trans. Robot. 34 , 645–659 (2018).

Song, Y., Romero, A., Müller, M., Koltun, V. & Scaramuzza, D. Reaching the limit in autonomous racing: optimal control versus reinforcement learning. Sci. Robot. (in the press).

Foehn, P. et al. Agilicious: open-source and open-hardware agile quadrotor for vision-based flight. Sci. Robot. 7 , eabl6259 (2022).

Jones, E. S. & Soatto, S. Visual-inertial navigation, mapping and localization: a scalable real-time causal approach. Int. J. Robot. Res. 30 , 407–430 (2011).

Download references

Acknowledgements

The authors thank A. Vanover, T. Bitmatta and M. Schaepper for accepting to race against Swift. The authors also thank C. Pfeiffer, T. Längle and A. Barden for their contributions to the organization of the race events and the drone hardware design. This work was supported by Intel’s Embodied AI Lab, the Swiss National Science Foundation (SNSF) through the National Centre of Competence in Research (NCCR) Robotics and the European Research Council (ERC) under grant agreement 864042 (AGILEFLIGHT).

Author information

Authors and affiliations.

Robotics and Perception Group, University of Zurich, Zürich, Switzerland

Elia Kaufmann, Leonard Bauersfeld, Antonio Loquercio & Davide Scaramuzza

Intel Labs, Munich, Germany

Matthias Müller

Intel Labs, Jackson, WY, USA

Vladlen Koltun

You can also search for this author in PubMed   Google Scholar

Contributions

E.K. formulated the main ideas, implemented the system, performed the experiments and data analysis and wrote the paper. L.B. contributed to the main ideas, the experiments, data analysis, paper writing and designed the graphical illustrations. A.L. formulated the main ideas and contributed to the experimental design, data analysis and paper writing. M.M. contributed to the experimental design, data analysis and paper writing. V.K. contributed to the main ideas, the experimental design, the analysis of experiments and paper writing. D.S. contributed to the main ideas, experimental design, analysis of experiments, paper writing and provided funding.

Corresponding author

Correspondence to Elia Kaufmann .

Ethics declarations

Competing interests.

The authors declare no competing interests.

Peer review

Peer review information.

Nature thanks Sunggoo Jung, Karime Pereida and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data figures and tables

Extended data fig. 1 residual models..

a , Visualization of the residual observation model and the residual dynamics model identified from real-world data. Black curves depict the residual observed in the real world and coloured lines show 100 sampled realizations of the residual observation model. Each plot depicts an entire race, that is, three laps. b , Predicted residual observation for a simulated rollout. Blue, ground-truth position provided by the simulator; orange, perturbed position generated by the Gaussian process residual.

Extended Data Fig. 2 Multi-iteration fine-tuning.

Rollout comparison after fine-tuning the policy for one iteration (blue) and two iterations (orange).

Supplementary information

Peer review file, rights and permissions.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and Permissions

About this article

Cite this article.

Kaufmann, E., Bauersfeld, L., Loquercio, A. et al. Champion-level drone racing using deep reinforcement learning. Nature 620 , 982–987 (2023). https://doi.org/10.1038/s41586-023-06419-4

Download citation

Received : 05 January 2023

Accepted : 10 July 2023

Published : 30 August 2023

Issue Date : 31 August 2023

DOI : https://doi.org/10.1038/s41586-023-06419-4

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

By submitting a comment you agree to abide by our Terms and Community Guidelines . If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.

Quick links

• Explore articles by subject
• Guide to authors
• Editorial policies

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.