PRISM turns 4 real videos into 256 counterfactual variants to train humanoid robot policies
berkeley_ai · x · 2026-10-06
Researchers at Amazon FAR (with Pieter Abbeel, Jitendra Malik, Karen Liu and others) introduce PRISM, a scalable real-to-sim-to-real pipeline for humanoid loco-manipulation. It expands 4 real interaction videos into 256 counterfactual variants in simulation, recovers robot-object trajectories, and trains a single policy that generalizes across diverse objects, scales, and spatial layouts. The work targets the long-standing bottleneck of finding suitable videos to teach robots at scale, and has been accepted to CoRL'26.
More from Embodied
- $3,500 local AI PC under fire: only 24GB VRAM and priced below its own parts cost — BLUECOW009 · 2026-10-06
- Parts alone cost ~$4,500 but Ghost's Core AI PC sells for $3,499 — where's the catch? — BLUECOW009 · 2026-10-06
- Actual raw lidar image from Waymo's 6th-gen sensor suite shared online — reed · 2026-10-06
- RT-SAFE benchmark: 94.1% of 8 frontier VLMs reach goals, only 0.7% finish with zero safety events — Lianhuiq · 2026-10-06
- Vibe Robotics: Artifact Arena makes frontier models engineer competing robots — kaixhin · 2026-10-06
- PointWAM: 3D world action model beats dexterous manipulation SOTA by 11.7 points — Chunghyun Park · 2026-10-06