RoboPapers

Starting to listen to these podcasts hosted by Chris Paxton. I think if you just spend the time to read those papers, then you can come up with some brilliant ideas.

Episode 1: SAM2Act

• Why not just increase the context window size so that the robot remembers more things, as opposed to coming up with new architecture? Author talks about compounding error, I don’t understand what he refers to that

Episode 2: Robot Utility Models

Episode 4: Vision Language Models are InContext Value Learners by Jason

Episode 6: FP3 A 3D Foundation Policy for Robotic Manipulation

Episode 15: Navigation World Models

Episode 17: EgoZero Robot Learning from Smart Glasses

Episode 18: HuB Learning Extreme Humanoid Balance

• Chris paxton: “I think these were some of the coolest humanoid balancing videos i’ve seen”

Episode 19: Learning to Drive from a World Model

Episode 20: VideoMimic

Episode 22: DexWild

• sheesh my boy jason liu here