Society of Robots
Search and Index Search Here

 Parts List
 Robot Forum
 Member Pages
 Axon MCU
 Robot Books

 How To Build
  A Robot




 Robot Journals
 Robot Theory

    How to Write
    a Science Paper

    As of this writing I've been a robotics researcher for the last ~10 years, and have published over a dozen papers. I'm also occasionally an invited reviewer for research papers. For a beginner, the process of writing a science paper can be confusing and maybe even overwhelming. I'd like to share my advice on how to write your research paper in a way to get it easily published for journals and conferences. Although I use the field of robotics for my point of view and examples, the information applies to any engineering field.

    But before I go into how to write a paper, I will explain how to *not* write a paper: a science paper is not a school report. Do not write a paper that goes something like this,

      "I made a cool robot. Here are some pictures of it. We had X problems with it. Then we tested it twice and it was awesome."

    Some of you are also only writing a paper to boost your resume, or to get into grad school. But - a science paper is about advancing science. It is not about writing a report or advancing your career. If you didn't do good science, you are doing a disservice to science by submitting a poorly researched paper.

    Note: The robotics field is overwhelmed by 'I made a cool robot' papers with little to no science involved. It's a pain to sort through them to find actual real science papers. Perhaps you are concerned about 'publish or perish', but don't choose quantity over quality to the detriment of science.

    Step 1:
    Before even writing your paper, decide on a scientific problem that needs solving to spend the next year or two on. I did not say 'decide on a cool robot to make', and if you have already, you've got it backwards. Too many researchers want to make a 'cool robot', and then spend time solving the required engineering problems. Instead you should be theoretically solving scientific questions, and only later build robots to experimentally prove your theoretical solutions to those questions.

    Step 2:
    Do a literature review. Has the problem already been solved, or at least partly solved? What theories have been proposed by other researchers? Are there any other research papers that can give you good ideas? If you are an engineer building a fish robot, would that marine biology paper describing an anatomical fish dissection be useful? Google Scholar is an incredibly useful tool - give it a try. I even have it set up to alert me the instant new relevant papers are published.

    Step 3:
    At what point during research should one start writing the paper?

    You should start on the same day you begin your research. Write down the problem you are solving, and why the world (or at least other researchers) would benefit with a solution to this problem. Theoretically solve the problem with math and thought experiments. Design some physical experiments to prove it, perform the experiments and collect data, and then use that data to draw conclusions. When you're done, just copy/paste all that into your paper - this is a good 80% of all you need to write!

    Step 4:
    After completing your research, find an appropriate conference or journal to publish in. Download their example paper format document (usually a .doc file), and start replacing text directly in it with your own.

    Step 5:
    Write the paper.

    === TITLE, AUTHORS ===

    This is the header of your paper. Select an informative paper title, such as 'How Leg Length Affects Biped Balance wrt Running Speed.' Don't write a paper titled, 'C.O.O.L. Robot: Initial Design and Construction' - that's just a progress report and likely lacks any real science. Some authors write 'Design and Construction' papers for robots that don't even work yet - don't do that. Finish the robot, run some tests, then write your paper. 'Initial Results' type papers are acceptable for conferences.

    The Author section is political. The person who physically wrote the paper is always 1st author. Those who put the next amounts of effort are 2nd and 3rd authors, etc. Your guiding professor, if he/she put real effort into guiding the team, gets listed as the last author.

    The place of research should be your university, workplace, and/or whoever funded your research.

    Include contact info for at least two authors using emails which will be used for at least the next 5 years. Many students like to use a university email address only to have it disabled a year later after graduation. You could be contacted by others looking to get clarifications with your paper, or requesting advice for similar research.

    Use your full name, including middle initial. If you have a common name, like John Smith, or Wu Wang, do your best to include even more identifying information. There are already thousands of other published authors named J. Smith and W. Wang.

    Note: I find it a pain to track down authors who give poor contact info. Do they not want credit for their own work?

    === ABSTRACT ===

    The Abstract and Conclusion sections of papers are the most important parts of your paper, but also the hardest to write, too. You may be given advice to write your paper first, then come back to write the Abstract last. This is very bad advice! The Abstract is the most fundamental description and purpose of your research. If you can't describe your research and/or don't know the purpose of it, it will clearly show in your writing.

    Let's say researcher Mr. A is building a biped, but he isn't sure how long the legs need to be for optimal running speeds. So he goes to Google Scholar and searches for papers that have already solved the problem. He will only read the paper Title and the Abstract, and maybe the Conclusion. He wants to find an abstract that says something like,

      "Abstract - The relationship of biped leg length is experimentally compared to biped robot speed. A mathematical equation is presented to calculate optimal leg dimension parameters for use in any biped leg design."

    Knowing your paper is relevant, he will read it and go on to finish his walking robot. But Mr. A will probably pass on your paper if the abstract instead reads as,

      "Abstract - We built a robot leg. It ran at 2mph. This paper documents the design."

    Why? Because Mr. A wants his biped to run at 1mph. Or 10mph. Or that leg design is too weak for his robot. But if he had a set of equations to use, he can then apply it to any size robot or running speed. When writing your abstract - and actually the whole paper - you should always be thinking, 'how can someone else use this information?'

    If people don't find your paper useful, then they won't use it. Or reference it in their research papers. No?

    For most papers, an Abstract only needs three sentences. The abstract should follow this format:

    We solved the problem of _______. This was done with method ________. It was determined that _________.

    For example,

      "Abstract - We solved the problem of how leg length affects biped running speed. This was done using both simulation and with an experimental biped which has extendable legs. It was determined that optimal leg length decreases by the square of the running speed."


    Note: It should take only minutes for you to write a good abstract. If you have trouble writing your abstract, this could be an unfortunate sign that you were directionless in your research. Talk with your team leader about it before going forward.

    === KEYWORDS ===

    When writing your ~five or so keywords, imagine someone using Google to search for your paper. What are the five keywords a person would most likely type to find your paper? The journal or conference may require you to only use their list of acceptable keywords. This is to help them classify your paper within their system. Following my robot leg example from above, I'd do something like:

      "keywords - biped, leg length, running speed, robot"

    === INTRODUCTION ===

    The introduction is where you make the case that your paper is important and solves some major unsolved problem. Start off by summarizing the current state of research. Then state there was this missing piece of information, and why it is important that it gets solved. Finally, state how your paper solves, or attempts to solve, that problem. Example,

      "Introduction - Robots that can walk on two legs is the holy grail of robotics. Building running robots involves solving many highly complex problems, with many interacting variables. One such unknown is how leg length affects running speed. If this problem can be solved, biped robots can achieve higher running speeds at higher efficiencies. Previous researcher Mr. X solved the problem for walking, but our work showed that it was not valid for running. We document in this paper a proposed solution to the speed vs. leg length relationship."

    My example is a bit oversimplified. Your research is likely to be more specialized and much more complex, with a ton of previous research involved, so your intro might take up half a page.

    === BACKGROUND ===

    The Background section is usually not necessary as your Introduction should cover the background already. But let's say you've been working on this problem for many years and have a dozen papers already published on the subject. For a reader to understand your current paper, he/she would have to read the dozen other papers first - a lot of reading! Instead, you can quickly summarize and self-reference your previous works to minimize any additional required reading. The Background section will make your paper 'self contained'.

    === THEORY ===

    Explain succinctly what your theory is, and how the following results will prove it. This section is to prepare the reader for the following experimental section. I rarely use a separate theory section, instead preferring to place it at the end of my Introduction.

    === Experimental Setup ===

    Explain how your experiment is set up, and what specifically you are measuring. Use CAD images to show mechanical designs, and little bubble text and arrows to explain the various parts. Explicitly specify what components you are using, the manufacturer, materials or chemicals you used, name the sensors you have, etc. Give dimensions, weight, and power requirements of your robots. State the microcontroller you used, and where the reader can access your software (if open sourced). Your reader should have enough information to be able to copy your experiment. If not, then how can your work be useful to him/her?

    For example,

      "To test biped running speed, a 10 meter flat track using standard model F1329 type carpet was layed out. A laser trip wire (model #AUH2398KJH) was set up at starting and ending positions. The 10kg biped was allowed to reach maximum speed before tripping the first laser wire, which triggered the start timer. The end timer was recorded as the biped tripped the second laser wire. Three trials were performed at each robot speed.

      DC servomotors (Hitec AS1234) were run at 7V to power each leg joint. A special extending mechanism allowed for lengthening of leg joints for each experiment set. See Figure 3 for a mechanism diagram."

    Some authors like to make a 'specifications' chart which quickly lists out important values, such as weight, maximum speed, power consumption, dimensions, etc.

    === RESULTS ===

    This is where your data goes. Use charts with pretty colors, add in trend lines to data plots, and pretty simulation diagrams. Next to all data explain in words what the reader is looking at, and how that result is significant to your theory.

    Don't forget to label the X and Y axis of every graph with units! If you have equations, specify in English what each variable and Greek symbol represents. For example, is 'd' for distance - or for 'diameter'?

    For example,

      Equation (1) represents the relationship of leg length to running speed of a robot. L1 is lower leg length, L2 is upper leg length, v is velocity, and M is robot mass.

      (1) (L1 + L2) * 2 * M = v

    (I made that equation up)

    Be ethical with your results. Don't intentionally leave out data that doesn't quite fit your theory. Don't fudge the data to fit curves. It sucks to spend a year collecting data to prove your theory, and then have data that doesn't fit the theory days before a deadline. But this is the reality of science - sometimes a theory is wrong, accept it and start over. Or publish it as a negative result.

    Note: Negative results are generally difficult to get published, but sometimes still very informative and useful for other researchers. It's like saying, "this idea looks good, but really it isn't so don't waste your time trying it."

    If you are dishonest with your data - and you are caught - your credibility as a researcher will be forever doubted. You will be expelled from your lab, your stipend confiscated, and your research career will end. Or even worse, if you aren't caught, your data can damage science for years. Poor/faked medical research can even result in deaths.

    === DISCUSSION ===

    The discussion section is where you write down your applicable thoughts to the paper. The discussion section is more of a 'what if' as opposed to hard facts. It's where what can't or wasn't proven in your research gets written. And where you discuss the various issues you had, and why you avoided certain types of experiments.

    Invariably there will always be caveats in your research. Maybe your experimental setup had physical or financial limitations and so you weren't able to study all aspects of the problem or at a sufficient level of accuracy. Perhaps after doing the research you determined a set of conditions which your results may or may not apply, so you suggest further research is needed to address those concerns.

    And you can discuss what your team is currently working on for the next paper.

    For example,

      "Discussion - Although our above research shows definitively the optimal robot leg length for any biped running speed, we did not factor in terrain in our calculations. Our experiments were only performed over smooth flat terrain, and not over rougher surfaces such as carpets or hilly areas. Also, our XYZ speed sensor was experimentally determined to have a ~15% error. Despite this, the results are still significant and valid although our equation constants may be slightly off. Our team is currently working on addressing the question of how terrain affects running speed for a future publication."

    === CONCLUSION ===

    Restate your conclusions succinctly, list what you've proven or disproven, etc. The conclusion isn't always necessary, as quite often it just restates what's already written in the abstract. Your paper may also justifiably be lacking a concrete conclusion - for example if your paper only demonstrates the performance of an algorithm without a concluding theory.

    DO NOT include new facts or information into the Conclusion that hasn't already been stated previously in the paper.

    The Conclusion formula is:
    The problem of _______ was resolved using the method ________. It was determined that _________.

    For example,

      "Conclusion - The problem of how leg length affects biped running speed was resolved using both simulation and with an experimental biped which has extendable legs. It was determined that optimal leg length decreases by the square of the running speed while on flat terrain."


    Write a paragraph thanking all those who gave you significant advice, helped you with setting up experiments, machined parts for you, etc. These people were important to your research, but weren't involved enough to be authors.

    For example,

      "Acknowledgements - The authors would like the thank Mr. C for his significant help in setting up and debugging Y experiment. We also thank Ms. R for assistance in 3D printing the robot leg joints."

    === REFERENCES ===

    References are your way of showing your research is well versed and informed of the relevant issues. A paper lacking relevant references, or that fails to point out very similar research by others, is a red flag for a poorly written paper.

    For example, let's say someone else already published a paper comparing leg length to biped running speed for short robots. I'd expect the current author to at least compare his results to that paper, and to explain why his results could be better.

    This does not mean fill your paper with barely-relevant references just to claim 'I have many references'. There is no requirement to have many references. I've seen good papers with only 10, and I've seen some with over 100. It really depends on the topic studied.

    You should already have all your references ready from Step #2 at the beginning of this article. If you waited until after you finished your research to find references for your paper, this is a bad sign of being uninformed of previous research. A 'literature review' should be done long before your first experiments.

    See what people have already done. Perhaps they've already solved certain parts of your problem. You might also get really good ideas that can speed up your own work. Go over to Google Scholar and do a search, and download all the relevant pdfs. You'll probably get a collection of ~50+ papers, but you'll quickly narrow them down after only reading the abstracts and conclusions.

    Many papers will also be behind a paywall. If you don't want to pay $25 for a paper, just email the author for a copy. Some researchers publish their papers on their website for download.

    To add the references to your paper, you must follow the required format of the journal or conference you are submitting to. Some formats require listing in order of appearance in the paper. Others require listing by alphabetical order of the first authors. In my opinion each has advantages and disadvantages, but I usually choose the IEEE format if I can. Below is an example of the IEEE format, just one of many.

      [1] Jason Geder, John S. Palmisano, Ravi Ramamurti, William Sandberg, Banahalli Ratna, "A New Hybrid Approach to Dynamic Modeling and Control Design for a Pectoral Fin Propelled UUV," 15th International Symposium on Unmanned Untethered Submersible Technology, Durham, New Hampshire, USA on August 19th-22nd, 2007.

      [2] John S. Palmisano, Ravi Ramamurti, Kerr-Jia Lu, Jonah Cohen, William Sandberg, Banahalli Ratna, "Design of a Biomimetic Controlled-Curvature Robotic Pectoral Fin," IEEE International Conference on Robotics and Automation, Rome, Italy, 2007.

      [3] etc. etc.

    Step 6:
    Spell check and grammar check! Copy/paste into Microsoft Word to catch 95% of the errors. If you aren't a native English speaker, find a native speaker to read over it. Don't submit until you do, because incomprehensible grammar results in papers getting rejected.

    Step 7:
    After your co-authors review your paper and give a thumbs up, print your paper to PDF and submit. I use PrimoPDF to convert Word files to PDF.

    You can set up your own page with Google Scholar to track down those referencing your own papers. It's not that accurate, missing the occasional reference, but still very useful. It also gives you the alert capabilities for tracking new publications.

    This is an example of my papers on Google Scholar: John's publications

    Another question you may have is 'how long should the paper be'? It really depends on your research, and on the requirements of the conference and journal. Generally aim for between 5 and 8 pages. Try to be succinct but thorough. As my 7th grade English teacher would say, "Start at the beginning and finish at the end." Or as Einstein advises, "if you can't explain it simply, you don't understand it well enough."

    For those who are more familiar with publishing science papers, you are probably aware of the serious drawbacks to the current journal system. I've been thinking about it for many years and at some point I'll write about it.

Get Your Ad Here

Has this site helped you with your robot? Give us credit - link back, and help others in the forums!
Society of Robots copyright 2005-2014
forum SMF post simple machines