PRISM Trains Humanoid Robots with Counterfactual Video Data
Amazon FAR, UC Berkeley, Stanford and CMU released PRISM (CoRL 2026), a real-to-sim-to-real framework that generates 256 counterfactual variants from 4 real videos to train humanoid loco-manipulation skills with zero real-robot fine-tuning.
2026-10-01 ~ 2026-10-01 · 2 related posts
- Amazon FAR's PRISM Turns 4 Real Videos Into 256 Counterfactual Variants to Train Humanoid Robots — chris_j_paxton · 2026-10-01
- PRISM scales humanoid loco-manipulation with counterfactual video generation, no real-world fine-tuning — Zihan Wang · 2026-10-01