KAIST's H2RB transfers human demonstrations to robots, boosting object transfer from 8.0 to 51.8
kaist-ai · hf · 2026-11-05
KAIST AI and RLWRLD present H2RB (Human-To-Robot Bridge, arXiv, Nov 3 2026), adapting pretrained VLA policies to new robots using unpaired human demonstrations, avoiding costly embodiment-specific data collection.
- Challenge: the human-robot embodiment gap spans both representation and action spaces, especially when the VLA was pretrained mainly on robot data
- Bridging both interfaces: geometry-anchored alignment grounds human and robot observations in a compatible representation; a decoupled-decoder tokenizer shares action latents without forcing a single action mapping
- Real-world results: object transfer jumps from 8.0 to 51.8, skill transfer reaches 62.7 where co-train baselines fail entirely; under object-placement shifts, OOD transfer improves from 6.0 to 45.3
- Implication: human demonstrations can serve as a scalable supervision source for adapting pretrained robot policies
More from Embodied
- Corebrum OS adds built-in VNC screen share, demoed remote-viewing an NVIDIA Thor robot from a Mac — chrismatthieu · 2026-11-05
- NerF Author Ben Mildenhall Teases Mystery AI Gadget With Robot Icon and 8:00 Display — XRarchitect · 2026-11-05
- New AI glasses UI concept demo shows off futuristic interface design — AnhPhuNguyen1 · 2026-11-05
- Berkeley project uses pretrained base policy's support to guide exploration for fast online adaptation — abhishekunique7 · 2026-11-05
- "Robotics simulation platform" startups are a dead end — agents already removed the need — carlosdponx · 2026-11-05
- Runtime (YC F26) debuts simulation-to-real robot policies, threads a rope with no fine-tuning — ycombinator · 2026-11-05