Weekly or more
RoboPapers
Hosted by Chris Paxton and Michael Cho
Chris Paxton and Michael Cho talk through new robot learning papers with the authors.
Episodes 3
Audio
Ep#106: Flex-π: A Multi-Stream World-Action Model with Compute Flexibility
The Flex-π team on a 6-billion-parameter world-action model that predicts 3D geometry and object features, not only color.
Audio
Ep#105: Beyond Imitation: Self-Improving Robot Policies via Off-Policy Q-Planning
Varun Giridhar and Animesh Garg on adding a learned value function so a large robot policy can improve from its own successes and failures.
Audio
Ep#104: Light-Loco-Parkour: Versatile Perceptive Whole-Body Locomotion via Multi-Skill Distillation
Xiaodao Chen and Yuntao Ma on a humanoid policy that decides on its own when to walk, vault, or climb using onboard sensing.