Manifold generative annealing improves robot contact motion safety
Manifold Generative Annealing for Contact Constraint-aware Impedance Control with Reinforcement Learning Prior
Robotics
Summary
Robots that interact with their environment face challenges when the way they touch or push objects changes unexpectedly. The authors propose a new method called Manifold Generative Annealing (MGA) that combines learning from past robot experiences with model-based simulations to create safer and more adaptable movements. This approach helps robots adjust in real-time to new contact conditions while respecting physical limits. Their tests included tasks like surface scanning and peg insertion, showing better safety and success compared to previous methods.
What this means in practice
- •For robotics engineers: Implement robust robot controllers that adapt impedance and motion safely under uncertain contact conditions during tasks like assembly or surface inspection.
- •For automation integrators: Enhance industrial robot systems to handle variable contact scenarios with fewer failures and improved safety margins using combined learning and model-based methods.
Authors
Zhilin He, Guoquan Qi, Changliu Liu
Abstract
Contact-rich robotic tasks require coordinated motion and impedance under uncertain interaction dynamics and physical constraints. While recent methods on learned policies provide useful interaction strategies, they may fail when contact conditions differ from those encountered during training. Model-based generative methods show promising receding-horizon trajectories optimization results, but often lack structured behavioral guidance and refinement that jointly accounts for constraint geometry and actual contact responses. To this end, we propose Manifold Generative Annealing (MGA), a joint model-free and model-based generative framework for motion--impedance optimization. MGA first organizes pretrained reinforcement learning policies into structured sequence proposals, retaining task-specific interaction strategies without committing to their execution. It then evaluates these proposals alongside exploratory samples through model-based rollouts for refinement or separate execution selection. Crucially, realization-aware manifold refinement aligns updates from the refinement bank with both task constraints and locally achievable physical responses, allowing learned experience to guide online adaptation under changing contact conditions. Experiments cover surface scanning, peg insertion, and humanoid pushing, with improved safety success rate of impedance control under nominal and sensing-shift conditions.