Reproduce “Physically Plausible Full-Body Hand-Object Interaction Synthesis” 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
Jona Braun, Sammy Christen, Muhammed Kocabas, Emre Aksan, Otmar Hilliges
A physics-based, hierarchical reinforcement learning framework synthesizes full-body human interactions, including dexterous object grasping and manipulation along specified trajectories. The method, developed at ETH Zurich, produces physically plausible motions with significantly reduced foot-skating and interpenetration, demonstrating higher grasp success rates and accurate trajectory following compared to kinematics-based baselines.
Key insights
- Decoupling the learning of full-body movements and dexterous hand control via hierarchical priors is crucial for synthesizing complex, multi-scale interactions and preventing mode collapse.
- Physics simulation, combined with reinforcement learning, can generate motions that are more physically plausible and free from artifacts than even ground-truth motion capture data, by explicitly enforcing physical constraints.
- Goal-conditioned policies and robust target guidance mechanisms significantly improve the ability to achieve stable grasps and follow diverse, user-defined manipulation trajectories.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.