
Episode 18: Oleh Rybkin, UPenn, on exploration and planning with world models
Generally Intelligent
How to Train a Biped Robbert to Walk in Simulation
For quadrupeds, you don't need vision. So you just need good physics. And we have good physics. We can actualy train right now with wen fors learning a robet. That's really good an tion. For bipeds, there's been a lot of papers that everybody just completely ignores cause they don't even know who's writing them. There was a paper from berkeley that did back flaps. I think thes dates are just the start, because they are actually trained in civilation. But maybe now that we have these robots, you can start collecting the data in the real world,. and you can come up with much better policies by collecting the data
00:00
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.