P-BFM: unsupervised RL lets Unitree G1 humanoids adapt to rough terrain from flat-ground references
ChongZzZhang · x · 2026-10-17
Researchers present P-BFM (Perceptive Behavioral Foundation Models), letting humanoid robots adapt to stairs, slopes, and rough terrain using only flat-ground motion references — no terrain-specific demonstrations. It uses unsupervised RL with forward-backward representations to encode a terrain-agnostic intent space, decoupling intent from execution, while depth perception adapts whole-body control. One frozen policy handles motion tracking, goal reaching, reward inference, and teleoperation on a Unitree G1, with an interactive browser demo, paper, and code released.
More from Embodied
- M5 Max MacBook Pro draws over 100W under load, stock Apple charger may not keep up — AaronBergman18 · 2026-10-17
- Standard Bots CEO: robots are at their '1973 moment'—and his plan to win US manufacturing — Rewkang · 2026-10-17
- Google Gemma hardware hackathon at AGI House offers $45k prizes, $15k for LiteRT track — agihouse_org · 2026-10-17
- Researcher builds generative machine vision pipeline reconstructing scenes from monocular video — RexDouglass · 2026-10-17
- Starkey Launches Omega AI+ Hearing Aid With Quad DNN Architecture and Sleep Tracking — BrandonSawalich · 2026-10-17
- RAX open-sources everything: zero demos, all experiment sessions included — du_yilun · 2026-10-17