Debate: Sim Distillation vs Scalable Human Data in Robot Learning

Kyle Morgenstein argues that distilling single-task policies trained in simulation fundamentally differs from behavior cloning on large-scale human data, which scales better. The debate also covers limits of learning general policies from human data, such as inferring complex motions like quadruped backflips.

2026-08-26 ~ 2026-08-26 · 2 related posts

• Robotics training debate: sim distillation vs. human behavior cloning — KyleMorgenstein · 2026-08-26
• Human Data Weakness: Can Quadrupeds Learn Backflips? — KyleMorgenstein · 2026-08-26