Guide

How to present a robotics paper in ten minutes

You said yes on Tuesday and the club meets on Saturday. This is the shape of a talk that holds up, and the order to build it in.

Build it backwards

Write the last slide first: one sentence a tired person could repeat correctly the next morning. If you cannot write that sentence, you have not finished reading. Everything before it exists to make that sentence land.

The six slides

1. The problem. What was broken before this paper, stated in terms of what a robot could not do. Not "prior methods are limited": say which capability was missing.
2. What came before. Two or three named approaches and the specific thing each one gets wrong. This is the slide that earns you the right to present the method.
3. The method. One figure from the paper, and the smallest possible description of what is new. Resist explaining the whole architecture; explain the part that would break if you removed it.
4. What they measured. The benchmark, the baseline, the number of trials or seeds, and the headline result. Say the numbers out loud, including the ones that did not improve.
5. Where it breaks. Assumptions, failure cases, and what the authors admit in the limitations section. If they admit nothing, that is itself the finding, and you should say so.
6. The takeaway. Your sentence from before. Then stop talking.

Say the assumptions before the results

The Utah robotics reading group makes this a rule: the presenter lists the paper's three to five core assumptions before discussing what it achieved. It works because a robotics result is almost always a result given a gripper, a scene distribution, a simulator, an observation space and a reset policy. State those first and the numbers arrive with their context attached, which is the difference between informing a room and impressing it.

A worked skeleton

Every paper page on this site carries a generated version of exactly this skeleton, extracted from the paper's own sections, nothing rewritten and nothing invented. It is a starting point that saves you the blank page, not a talk. Read the sections it pulled from before you say them out loud.

See the kit for “World Models”

The evening before

• Read the abstract, introduction, and limitations. In that order, and stop there on the first pass.
• Read the results section and write down the baseline and the trial count before you look at the headline number.
• Now read the method, with a specific question in mind rather than start to finish.
• Write the last slide. If it is hard, go back to the introduction and find the sentence where the authors state their claim.
• Pick two figures. Delete every other visual.
• Say it out loud once, on a timer. Whatever you cut at minute eleven was not load-bearing.

Common questions

How long should a paper presentation be?

Ten minutes of talk and ten of discussion is the format most open clubs use; lab seminars usually run 45 plus 15. If you only have one slot, choose ten. A ten-minute talk forces you to decide what the paper is actually claiming, and a 45-minute one lets you avoid that decision for 44 of them.

What should be on the slides?

Six slides: the problem, what came before, the method, what was measured, where it breaks, and the takeaway. One figure from the paper on the method slide and one results table on the measurement slide. Everything else is talking.

Do I need to understand the whole paper?

No, and pretending to is the most common way a talk goes wrong. Understand the claim, the setup that supports it and the conditions it was tested under. If a section defeats you, say which one and ask the room. That is the entire point of reading together.

How do I read a results table without being fooled?

Check what the baseline is, whether it was tuned as carefully as the method, how many seeds were run, and whether the headline number comes from the same conditions as the baseline number. In robotics, also check how many real-world trials there were: twenty trials on one robot is common and is not a strong result.

What if I have never presented before?

Present a paper you do not like. It is much easier to be interesting about work you have a specific objection to, and the room will engage with the objection rather than watching you narrate figures.