Integrated Task and Motion Planning for Safe Legged Navigation in Partially Observable Environments

This study proposes a hierarchically integrated framework for safe task and motion planning (TAMP) of bipedal locomotion in partially observable environment with dynamic obstacles uneven terrain. The high-level planner employs linear temporal logic reactive game synthesis between the robot its provides formal guarantee on navigation safety completion. To address environmental partial observability, belief abstraction model is designed by partitioning into multiple regions employed at to estimate obstacles' location. additional location information offered enables less conservative long-horizon actions beyond guaranteeing immediate collision avoidance. Accordingly, synthesized action sends set middle-level while incorporating specifications extracted from theorems based reduced-order (ROM) process. ROM design criteria sampling algorithm generate nonperiodic plans that accurately track actions. At low level, foot placement controller an angular-momentum inverted pendulum implemented ankle-actuated passivity-based full-body trajectory tracking. external perturbations, this also investigates sequential composition keyframe state achieves robust transitions against perturbations through reachability analysis. overall TAMP validated extensive simulations hardware experiments walking robots Cassie Digit Agility Robotics.