Track manipulation, locomotion and embodied systems.
Search papers, labs, and topics across Lattice.
This paper introduces Manifold Generative Annealing (MGA), a hybrid model-free and model-based generative framework for contact-rich robotic tasks, addressing limitations of pure RL and model-based methods in handling varied contact conditions. MGA leverages pretrained RL policies for structured proposals and refines them using model-based rollouts, crucially aligning updates with task constraints and locally achievable physical responses. Experimental results on surface scanning, peg insertion, and humanoid pushing demonstrate improved safety and success rates for impedance control, even under sensing shifts.
Learned robotic interaction strategies can be robustly adapted to novel contact conditions by organizing RL policies into generative sequence proposals and refining them with realization-aware manifold updates.
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.