Full of childlike wonder. Teaching robots manners. RL Lead @ Apptronik. UT Austin PhD candidate. Past: Boston Dynamics AI Institute, NASA JPL, MIT ‘20.
i really hate ai-generated research blogs/presentations. i have recently talked to some people who believe that writing papers and blogs is a waste of time, and that it gets in the way of research, but i would like to revisit the fact that *sharing results* is a part of the scientific process.
effective presentation of findings is critical in enabling others in the community to understand your methods and results. personally, i am still not satisfied with the quality of writing that the models produce, and would therefore plead for people to still go the last mile to ensure the writing you publish is of high quality.
There are many papers solving motion tracking nowadays and we take a step back and look at the motions to track and question:
Can legged robots imitate human motion when their kinodynamic properties demand 𝗱𝗶𝗳𝗳𝗲𝗿𝗲𝗻𝘁 𝘁𝗶𝗺𝗶𝗻𝗴?
To answer this, we are happy to share DTMR!
Can a laptop train a humanoid robot controller?
Training a humanoid motion controller used to mean firing up a cluster of NVIDIA GPUs. Now you can train in 22 minutes on a MacBook Pro.
Project page: https://t.co/xq56MxAD3L
it was only a matter of time. if you do have a bunch of GPUs though, I got training down to 27 seconds on 4x RTX 6000 Pro. the whole pipeline, including retargeting, takes under a minute.
Can a laptop train a humanoid robot controller?
Training a humanoid motion controller used to mean firing up a cluster of NVIDIA GPUs. Now you can train in 22 minutes on a MacBook Pro.
Project page: https://t.co/xq56MxAD3L
@carlosdponx basically none of the model based estimators can handle general body contact. right now your best bet is to learn a concurrent estimator in sim, but the methods for dealing with contact this paper uses certainly could be extended to other body parts as well
many labs use reduced observation spaces for their controllers because they don’t think about state estimation. this silently may be hurting your policy performance. try this!
In a collaboration with Oxford, SNU, and Pisa, we developed a great legged-robot state estimator, from proprioception only, by being careful about when and how often to use contact events. Check out the results (and code) at https://t.co/YarYXc4l6W
@ChenTessler depends what state you’re trying to recover and what you include in the historical observations. blindly adding history can also hurt you!
@YouJiacheng ask “what are the types of states that make pick and place?” and “near the goal, far from the goal, holding the object, not holding the object” are a pretty clear factorization
people kept asking me for “hot takes” at CoRL. I was sorry to disappoint, because what I came up with felt lukewarm:
RL is the only thing that has ever worked at scale.
exploration + adaptation. that’s it. everything else is a party trick.
RL is the ultimate bitter lesson.
We used to hike with our robot, controller in hand.
Now it hikes autonomously, 10 km down from the Matterhorn. 🏔️🐄
We published a new paper in Science Robotics:
📄 https://t.co/32z37SK4KQ
🌐 https://t.co/vmy3IbrpyP
@YouJiacheng it doesn’t require complex rewards or curricula much of the time! especially with motion tracking for locomotion or works like SGS or AlphaNist, we’re converging on much more general scalable setups
I’m fairly certain this x years left to contribute to the world intellectually has always been a psyop by big labs for their best and brightest agi pilled employees to keep them working longer hours and discourage them from going to competitors.
Take just a moment to consider what it means seriously.
Why 1-5 years? (I’ve been hearing 1-5 years for about 2 years now fwiw).
Why does contributing to the world intellectually even really matter to the extent that one must recenter one’s life around doing it in this window?
Admittedly I started gradschool knowing full well I was going into industry on day one, and I told my advisor that before I started. That’s a bit unusual I admit. I like to say I didn’t know what a PhD was and accidentally became a scientist. But I did fall in love with the process, unraveling mysteries, and experimental design. The process of walking through mentally who could you take something abstract and making it into a testable hypothesis.
What I will say is the problems don’t become more or less interesting because a computer can do them for you.
If you are working on something you dislike so much you wouldn’t do it if there was another person that was faster or better at it, it’s not the faster better persons fault … you just don’t actually like doing it. Maybe it’s just a prestige thing idk, but sounds like if it was the case then it was never the “intellectual contribution” or joy of problem solving you really cared about in the first place.
Don’t over think this shit. Just do what makes you happy. If AI solves that problem. Great, I promise there will be more out there waiting. Hell even if there is not this existential threat to your own lives meaning mindset isn’t helpful or healthy.