Reproduce “Contact-Safe Reinforcement Learning with ProMP Reparameterization and Energy Awareness” 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
Bingkun Huang, Yuhe Gong, Zewen Yang, Tianyu Ren, Luis Figueredo
The PPT (ProMP PPO Energy-Tank) framework combines Probabilistic Movement Primitives, Proximal Policy Optimization, and an energy-tank passivity layer to enhance contact-rich robotic manipulation. This integration enables robots to perform complex physical interaction tasks with improved safety and stability, achieving higher success rates and smoother motions with lower peak forces in both simulated and real-world box pushing and maze sliding tasks compared to step-wise reinforcement learning methods.
Key insights
- Combining structured trajectory representations (ProMPs) with reinforcement learning significantly improves trajectory smoothness (reduced jerk) and contact continuity in physical interaction tasks.
- A passivity-based energy-tank mechanism provides explicit safety guarantees by actively regulating energy flow, effectively preventing hazardous force or power bursts during contact.
- The integrated framework facilitates robust sim-to-real transfer and generalization to unseen contact geometries without requiring policy fine-tuning, demonstrating practical utility in uncertain environments.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.