Event
Stage-Conditioned Policy Optimization Enables Multi-Stage Mobile Manipulation
1 reports · 1 sources
Simone Manni and co-authors propose stage-conditioned policy optimization, extending PPO to accomplish multi-stage robotic mobile manipulation tasks with a single policy. The method introduces a stage-conditioned critic, stage-gated rewards, and a multi-stage reset curriculum, and adopts a decoupled hierarchical control architecture that separates low-level balance dynamics from high-level stage execution; the authors state that the policy can acquire complex mobile manipulation behaviors solely through environment interaction while incorporating high-level task structure directly into learning. The authors state that they have validated multi-stage mobile manipulation skills on real Unitree G1 and Boston Dynamics Spot equipped with a robotic arm, and have tested design choices through ablation experiments.
Report timeline
- Learning Multi-Phase Robot Loco-Manipulation via Phase-Conditioned Policy Optimization
The study proposes extending PPO into phase-conditioned policy optimization, using a phase-conditioned critic, phase-gated rewards, and a multi-phase reset curriculum to train a single policy for long-horizon loco-manipulation tasks. The method adopts a decoupled hierarchical control architecture that separates low-level balancing dynamics from high-level phase execution, and validates multi-stage skills on real hardware on the Unitree G1 and a Boston Dynamics Spot equipped with a manipulator arm, while testing design choices through ablation experiments.
arXiv cs.RO ·