korrents

What Chelsea Finn thinks about robotics

@chelsea-finn · 22 positions · 0 changes of mind

Assistant professor of computer science at Stanford, and co-founder of the robot foundation-model company Physical Intelligence.

Chelsea Finn did not write this page.

We collected these quotes from things they published elsewhere, and every quote links to where it was said. They have no account here and have not endorsed this site. Quotes are word for word; the short line under each one is our own restatement, not their wording. Their own site. Is this you? Claim it or ask us to remove it. Or tell us what is wrong here.

17 dated positions, 2026, in their own words. Our reading of what Chelsea Finn has said — not written or endorsed by them.

Our readingThe constraint returned to throughout is data rather than model size: diversity matters more than volume, low-quality data helps only when labelled as such, and most future data will come from attempts the robot makes itself. The reliability bar is thirteen hours of running, not one demonstration.

Their wordswe find that the performance on held out tasks decreases dramatically. Whereas if we um just take out a random 20% of the data that's less diverse than the most diverse subset, the performance um only decreases a little bit. And so this suggests that actually having really diverse data plays an important role in enabling it to generalize to new tasks.

It is the diversity of a robot data set, not its size, that produces generalization: dropping the most diverse slice hurts far more than dropping a random fifth.

  1. Our readingPhysical AI has to make far fewer mistakes than any AI deployed so far, because it is only useful when it acts autonomously on the world.

    Their wordsUh and this means that they're going to be far more useful when they're operating fully autonomously. And as a result, this requires us to develop physical AI systems that make far fewer mistakes than the machine learning systems that have been deployed thus far.

    ↗Chelsea Finn: This is the State of the Art in Roboticsyoutube.com 2nd of 22 in this recording

    Interesting

  3. Our readingHand-tuning a robot's data set cannot reach very high reliability, because people get tired; the system has to seek out its own missing data.

    Their wordsUm, and while this generally improves the reliability of the model, uh, people eventually get tired and it's hard to get really, really high reliability with a person that's manually tuning this. And so what would be even better is if the AI system itself can iterate on the scenario in which you want it to have higher reliability where it on its own automatically seeks out places where it needs more data, where it needs more supervision.

    ↗Chelsea Finn: This is the State of the Art in Roboticsyoutube.com 3rd of 22 in this recording

    Interesting

  5. Our readingReinforcement learning cannot be scaled for robots the way it was for language models, because every attempt spends real robot hours instead of data-centre compute.

    Their wordsNow maybe this isn't completely out of the question but this would be quite challenging uh to do and that's because the calculus is a little bit different. We're not just running compute to optimize for a use case. We're actually running the robot in the real world and using the hardware and attempting the task in the real world.

    ↗Chelsea Finn: This is the State of the Art in Roboticsyoutube.com 4th of 22 in this recording

    reinforcement learning

    Interesting

  7. Our readingA robot going down a wrong path is not producing training data; the efficient move is for a human to intervene and show the recovery.

    Their wordsand trying to fold two boxes together isn't useful data that will teach the model how to get better at the task. And so that would be kind of wasting a lot of time on the robot attempting to go down the wrong path for solving the problem. And so instead of spending a lot of time trying to do that task, what we'll do is we'll actually have a human intervene and show the robot what to do and how to recover from that situation.

    ↗Chelsea Finn: This is the State of the Art in Roboticsyoutube.com 5th of 22 in this recording

    Interesting

  9. Our readingWhether a robot policy is any good is settled by running it for thirteen hours straight, not by whether it completes the task once.

    Their wordskind of going back to this reliability question, we took this policy and we ran it not just once, but we ran it for 13 hours straight. Uh and we basically wanted to evaluate is this policy not only good at making a latte once, but can it do so reliably to the extent that it would be needed to be useful in the real world?

    ↗Chelsea Finn: This is the State of the Art in Roboticsyoutube.com 6th of 22 in this recording

    Interesting

  11. Our readingMost state-of-the-art robot foundation models have no memory at all, and that is what stops them doing long multi-step tasks.

    Their wordsSo, you might be surprised to hear that most state-of-the-art foundation models for robotics have no memory or no context. They're just operating on the current sensor observations, the current camera readings, uh, and predicting actions based off of that.

    ↗Chelsea Finn: This is the State of the Art in Roboticsyoutube.com 7th of 22 in this recording

    Interesting

  13. Our readingA robot does not need video to remember the last hour: past a few seconds, memory can be a text summary of what happened.

    Their wordsAnd then for longer memory, for memory that spans multiple minutes or multiple hours, we don't necessarily need video of exactly what happened in that past uh in that past history.

    ↗Chelsea Finn: This is the State of the Art in Roboticsyoutube.com 8th of 22 in this recording

    Interesting

  15. Our readingOne pre-trained robot model now matches or beats the specialists that were fine-tuned with reinforcement learning for the very tasks they were built for.

    Their wordswe see that the across the board the single PIO like pre-trained PIO7 model matches or outperforms the fine-tuned specialists that were developed with reinforcement learning post-training for those downstream tasks.

    ↗Chelsea Finn: This is the State of the Art in Roboticsyoutube.com 11th of 22 in this recording

    scaling lawsreinforcement learning

    Interesting

  17. Our readingIt is the diversity of a robot data set, not its size, that produces generalization: dropping the most diverse slice hurts far more than dropping a random fifth.

    Their wordswe find that the performance on held out tasks decreases dramatically. Whereas if we um just take out a random 20% of the data that's less diverse than the most diverse subset, the performance um only decreases a little bit. And so this suggests that actually having really diverse data plays an important role in enabling it to generalize to new tasks.

    ↗Chelsea Finn: This is the State of the Art in Roboticsyoutube.com 12th of 22 in this recording

    Interesting

  19. Our readingLow-quality data makes a robot model worse unless you tell the model the data is low quality; labelled as such, the same data makes it better.

    Their wordswithout metadata prompting when you add lower quality data from 80% data to 100% data the performance actually decreases which is perhaps not too surprising because you're adding lowquality data to your data mixture whereas with the metadata prompting the performance actually increases when you add that lowquality data

    ↗Chelsea Finn: This is the State of the Art in Roboticsyoutube.com 13th of 22 in this recording

    Interesting

  21. Our readingRobotics will not get a ChatGPT moment, because the distribution channel for a physical model is a physical robot and that cannot spread in five days.

    Their wordsI think that the distribution channel for physical models is going to be slower uh unfortunately because you actually need a physical robot there

    ↗Chelsea Finn: This is the State of the Art in Roboticsyoutube.com 14th of 22 in this recording

    OpenAI

    Interesting

  23. Our readingRobot models will reach ChatGPT-level capability within the next few years, even without a ChatGPT-style moment.

    Their wordsat the same time in terms of the capabilities of these models I think that we are really starting to get to the point where these models are actually useful in the real world and I think that getting to the kind of the capabilities of chat GBT I think is um yeah very much on the horizon in the next few years.

    ↗Chelsea Finn: This is the State of the Art in Roboticsyoutube.com 15th of 22 in this recording

    OpenAI

    Interesting

  25. Our readingMost future robotics training data will be robots' own autonomous attempts, the way most language-model data is now generated by the model itself.

    Their wordsUh and just like how we see in language models how now a lot of time is spent actually generating data, generating synthetic data by actually running the model and having it think through things. I think a lot of the data in the future in robotics is going to be the robot attempting to do lots of tasks in lots of real world circumstances.

    ↗Chelsea Finn: This is the State of the Art in Roboticsyoutube.com 18th of 22 in this recording

    Interesting

  27. Our readingWatching humans is not enough for a robot: there is no substitute for experience gathered on its own body.

    Their wordsUh and likewise um robots can't like watch a person doing something and then figure out how to do it themselves directly. They really need their experience on their own platform um to learn effectively.

    ↗Chelsea Finn: This is the State of the Art in Roboticsyoutube.com 19th of 22 in this recording

    Interesting

  29. Our readingLetting a robot imagine the next image helps, but it is not essential — the model was surprisingly good without it.

    Their wordsuh the we found that that leads to improvement and we saw in the shirt folding example we saw like a quantitative bump from using that sort of imagination compared to not using it. At the same time I think that the model actually performed surprisingly well without that as well.

    ↗Chelsea Finn: This is the State of the Art in Roboticsyoutube.com 20th of 22 in this recording

    Interesting

  31. Our readingRobots are slow because the people teleoperating their training data were slow, so the only two ways out are faster data or a policy faster than its data.

    Their wordsand yeah, I think it's either you need to figure out how to make the data faster or you need to figure out how to be faster than the data. We've seen the evidence of being able to be a little bit faster than the data.

    ↗Chelsea Finn: This is the State of the Art in Roboticsyoutube.com 21st of 22 in this recording

    Interesting

  33. Our readingRobot policies already show emergent capability: one transferred a skill from its right hand to its left with no such example anywhere in its training data.

    Their wordsUm the robot essentially had learned this sort of equivariance between his left hand and his right hand so that it could actually transfer uh behaviors from one hand to another. uh despite the fact that that was never in the data.

    ↗Chelsea Finn: This is the State of the Art in Roboticsyoutube.com 22nd of 22 in this recording

    Interesting