Harvard's GRASP refines human demos in physics sim for dexterous hands and Unitree G1
hankyang94 · x · 2026-11-01
Harvard's Computational Robotics Group introduces GRASP (Gradient-Guided Sampling for Contact-Rich Physics-Based Motion Retargeting) for dexterous hands and humanoids.
- Problem: kinematically retargeted human demos can look right while requiring contacts and dynamics the robot can't realize, forcing downstream RL to track physically inconsistent references.
- GRASP refines these references in physics simulation before policy learning, helping RL learn contact-rich motion tracking — demonstrated in pen spinning and humanoid–object interaction.
- Real hardware: open-loop trajectory replay on Sharpa hands and learned tracking policies on a Unitree G1.
- Paper, code, and dataset are open-sourced, aiming to scale motion tracking to more diverse contact-rich tasks.
Related event: Harvard Unveils GRASP for Contact-Rich Motion Retargeting(2 posts)→
More from Embodied
- Naval predicts 4-passenger drones will eventually end commercial air travel, with a long interregnum ahead — naval · 2026-11-01
- Meta Quest game scans your real room and turns it into a haunted house — Scobleizer · 2026-11-01
- RoboAuto: Robot Program Search With Multi-Level Physical Understanding — DJiafei · 2026-11-01
- Developer dresses robot 'Bob' as a ghost to hand out Halloween candy — chrismatthieu · 2026-11-01
- GRASP authors compare with SuperMimic: extracting more optimization signal from sim experience — hankyang94 · 2026-11-01
- Neuralink user designs 3D-printed parts with thoughts alone, holds them hours later — Scobleizer · 2026-11-01