GRASP: Gradient-Guided Sampling for Contact-Rich Physics-Based Motion Retargeting
GRASP: Gradient-Guided Sampling for Contact-Rich Physics-Based Motion Retargeting
This research presents GRASP (Gradient Recovery from Archived Samples for Proposals), a method for physics-based motion retargeting that combines sampling exploration with gradient-guided refinement without needing a differentiable simulator.
The paper introduces GRASP, a sampling-based optimization method that uses archived samples and gradient guidance to retarget human demonstrations for robots without requiring a differentiable simulator, validated on both simulated dexterous hands/humanoids and real hardware.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
demonstration count
1,352
View original evidence
eceding-horizon planning. Experiments on dexterous hands and humanoids show accurate retargeting with a shared tracking objective and no task-specific contact rewards, including 92.8% success on 1352 bimanual OakInk2 demonstrations. The refined references improve PPO tracking in pen spinning and humanoid-object interaction. We also demonstrate dexterous manipulation and whole-body humanoid motion on real hardware.
Open source S4
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
eceding-horizon planning. Experiments on dexterous hands and humanoids show accurate retargeting with a shared tracking objective and no task-specific contact rewards, including 92.8% success on 1352 bimanual OakInk2 demonstrations. The refined references improve PPO tracking in pen spinning and humanoid-object interaction. We also demonstrate dexterous manipulation and whole-body humanoid motion on real hardware.
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org