EPFL's TERRA pipeline teaches muscle-actuated agents to walk 9.4 hours of diverse terrain
amathislab · hf · 2026-10-01
EPFL researchers present TERRA, an end-to-end pipeline for terrain-aware retargeting and control of musculoskeletal locomotion.
- Problem: muscle-actuated RL agents can reproduce complex motions but remain confined to flat ground due to lack of aligned terrain geometry in motion datasets.
- Method: from kinematic trajectories alone, TERRA combines terrain priors, contact estimation, and negative free-space evidence to recover support geometry, respecting anatomical, tendon-continuity, and contact constraints during retargeting.
- Results: using motion-terrain pairs from five datasets, they train a single muscle-actuated policy on 9.4 hours of diverse locomotion, improving terrain accuracy, sharply reducing violations, and achieving the highest completion rates.
Project page: cnai.epfl.ch/terra
More from Embodied
- miHoYo unveils Gamer, a generalist AI agent that plays games by watching the screen — minchoi · 2026-11-17
- Korean J&J Tech builds pharmacy humanoid on Unitree R1 for shelf checks — CyberRobooo · 2026-11-17
- Maker Builds a Vacuum Pickup Screwdriver as a Tool for a Robotic Agent — ihorbeaver · 2026-11-17
- Claude designed three working devices; the 25-degree board tilt detail stands out — nukebuddy · 2026-11-17
- Vision Pro app Passage to drop subscription for one-time purchase on Dec 1 — wattmaller1 · 2026-11-17
- First Cybercab ride in Austin: reviewer says it beats Waymo handily — Scobleizer · 2026-11-17