Reproduce “ForceMimic: Force-Centric Imitation Learning with Force-Motion Capture System for Contact-Rich Manipulation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Wenhai Liu, Junbo Wang, Yiming Wang, Weiming Wang, Cewu Lu
Researchers at Shanghai Jiao Tong University developed ForceMimic, an integrated system comprising ForceCapture for natural human force-motion data collection and HybridIL, a diffusion-based policy for hybrid force-position control. This approach enabled robots to achieve 100% motion correct and 85% continuous peel success on the challenging zucchini peeling task, maintaining consistent interaction forces.
Key insights
- Explicitly learning and controlling force (via wrench prediction and hybrid control) is crucial for robust contact-rich manipulation, outperforming purely pose-based or implicitly force-aware methods.
- Natural, robot-free human demonstration collection with integrated force sensing significantly improves data quality and collection efficiency compared to traditional teleoperation.
- Simply inputting robot force sensing as an observation to a policy is not sufficient; the policy needs to learn how to effectively use this information, and the control strategy must align with the learned force predictions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.