Not All Diversity Helps: Noisy Robot Demos Waste Policy Capacity
chris_j_paxton · x · 2026-11-19
Robotics researcher Chris Paxton echoes Firat Candemir: with current imitation learning methods, "more diverse is better" doesn't hold. If every operator does the same task differently, the policy spends its capacity modeling that noise instead of the actual work.
Useful diversity comes from different workplaces and objects — not five slightly different grips on the same pick. Consistent demos beat a pile of slightly different takes.
More from Embodied
- NVIDIA Ships VSS 3.3 and Cosmos to Accelerate Visual AI Agents on Jetson — NVIDIAAI · 2026-11-19
- Tianji's wheeled humanoid Gento Luna hits 99%+ picking success in real retail — CyberRobooo · 2026-11-19
- France reportedly lobbying EU states to vote against Tesla FSD approval, FOIA reveals confidential exchanges — mitchdeg · 2026-11-19
- Solo builder shows off fully self-built quadruped robot, from firmware to actuators — wavefnx · 2026-11-19
- DIY $250 AI assistant device writes its own apps on a round touchscreen — 5im_Only · 2026-11-19
- 103M Gaussians run natively and smoothly on iPhone 15 Pro, no LOD needed — janusch_patas · 2026-11-19