There’s a lot of awesome research about LLM reasoning right now. But how is  learning in the physical world 🤖different than in language 📚?. In a new paper, show that imitation learning in continuous spaces can be exponentially harder than for discrete state spaces, even when

Very cool! In addition to optimizing inference-time search as a learning desideratum, this really speaks to power of building reward models purely from expert trajectories, via discriminative objectives. Excited to see how far this can go!.

RT @GuanyaShi: I am giving a talk "From Sim2Real 1.0 to 4.0 for Humanoid Whole-Body Control and Loco-Manipulation" at the RoboLetics 2.0 wo….

RT @NicholasEPfaff: Want to scale robot data with simulation, but don’t know how to get large numbers of realistic, diverse, and task-relev….

RT @canondetortugas: RL and post-training play a central role in giving language models advanced reasoning capabilities, but many algorithm….

RT @maiamindel: The Chicago Pope implies the existence of a New Keynesian Pope and a Behavioral Pope.

RT @nmboffi: really excited about this one -- please submit your best work, and come join us in beautiful Lyon to talk about machine learni….

RT @CMU_Robotics: Congrats to Andrea Bajcsy (@andrea_bajcsy) on receiving the NSF CAREER award! 👏 . Her work: “Formalizing Open World Safet….

RT @aleks_madry: Building AI systems is now a fragmented process spanning multiple organizations & entities. In new work (w/ @aspenkhopkin….

Update: 2pm is the likely time.

Please check my account for updates on the actual time ! Unfortunately, I was advised not to post a link here, but if you google “RL Theory Virtual Seminar”, and click “Next Seminar” on the sidebar, there will be registration form to view the talk through a Google meet.

Regarding time: The seminar is listed both at 6pm UTC (which would be 2pm ET) and 1pm ET; these seem to be off my a daylight savings time adjustment. We are actively in contact with the organizers to resolve the issue. .

Hey Everyone!! I will be giving a lecture at the RL Theory Virtual Seminar tomorrow, on my new paper about the “Pitfalls of Imitation Learning" in continuous action spaces. 🧵 below; please read because the time is somewhat TBD. 🧐

RT @aviral_kumar2: Before the (exciting) workshops on Sun, catch Vincent’s oral talk at the #ICLR2025 main conference on this paper today a….

RT @LerrelPinto: So excited for this!!! . The key technical breakthrough here is that we can control joints and fingertips of the robot **w….

Check out our paper to find out more, or join in on the RL theory seminar on April 29th at 1pm ET (follow for more details).

What does make a difference is using a richer policy parametrization! We give evidence that even for imitating this smooth, deterministic - i.e, unimodal - expert, Diffusion and action chunking have surprising benefits!

This result holds for imitating a smooth, deterministic expert demonstrator in a smooth, exponentially stable control system. Exponential compounding error can't be avoided by using a fancier learning algorithm - BC, Adversarial IL, and Offline RL all suffer.

RT @Princeton: The Porter Ogden Jacobus Fellowship, the University’s top honor for @PrincetonGrad students, is awarded to one Ph.D. student….

RT @abhishekunique7: So we did a bunch of projects with real world reinforcement learning - but it was often too inefficient to be practica….