CoRL 2026 paper GRA: synthetic robot videos should supervise geometry, not control
MikeShou1 · x · 2026-09-05
- 'Supervise What Survives' (arXiv 2606.24448) accepted to CoRL 2026, from Danze Chen et al. with Mike Zheng Shou
- Core insight — Asymmetric Preservation Principle: human-to-robot video generation preserves visible geometry (the 'where') while erasing control signals (the 'how'), so recovering pseudo-actions from synthetic pixels is a mismatched abstraction
- GRA extracts future 2D end-effector waypoints via pose estimation, retargeting, simulation and calibrated projection, feeding them through an auxiliary 2D head to supervise the VLA vision backbone; the action head trains only on real teleoperation demos
- Motivation: VLAs need massive video-action pairs but real teleoperation data is scarce; generated videos are a scalable, label-free alternative
More from Embodied
- XPeng commissions 80%-automated line for IRON humanoid, mass production by end-2026 — ingliguori · 2026-10-23
- HKU's TILDE Distills Embodied Foundation Models into Robot Policies — hkuhk · 2026-10-23
- 10 of 12 Beta Robots Worked Without Fine-Tuning — Do We Still Need It? — micoolcho · 2026-10-23
- EmbodiedEvo lets robot agents self-train, banking reusable Skills and Lessons across tasks — jiqizhixin · 2026-10-23
- KAIST's FRED lifts robot policy success from 19.4% to 50% via self-distilled future video representations — kaist-ai · 2026-10-23
- GALAX Launches World's First RTX 5080 With Dual 12V-2x6 Connectors at $3000 — chemist_slime · 2026-10-23