Human Data Weakness: Can Quadrupeds Learn Backflips?
KyleMorgenstein · x · 2026-08-26
Discusses the limitations of human data in training generalist policies, specifically how a quadruped might infer complex maneuvers like a backflips without generating specific data from single-task experts. Argues that distillation from sim-trained single-task policies differs fundamentally from large-scale behavior cloning with human data, as the former doesn't scale as easily.
Related event: Debate: Sim Distillation vs Scalable Human Data in Robot Learning(2 posts)→
More from Embodied
- Robotics firms profit by deploying for real customers over hype — kscottz · 2026-08-26
- 5 solenoid boards in 44 minutes: AI-assisted hardware development — debreuil · 2026-08-26
- Best Robot Fails From the World Humanoid Robot Games Go Viral — CSProfKGD · 2026-08-26
- Future delivery dude: humanoid robot demo — LinusEkenstam · 2026-08-26
- OOMWOO Open-Source Robot Vacuum: Raspberry Pi, ROS 2, and Fully Local Control — JeremyCMorgan · 2026-08-26
- Humanoid robots break sprint records but crash and burn at Beijing games — Ars Technica AI · 2026-08-26