Predictive Reachability for Embodiment Selection in Mobile Manipulation Behaviors

Mobile manipulators require coordinated control of navigation and manipulation to accomplish tasks. Typically, coordinated mobile manipulation behaviors involve base navigation to approach the goal followed by arm manipulation to reach the desired pose. The embodiment selection between the base and arm can be determined based on reachability. Previous methods evaluate reachability by computing inverse kinematics and activate arm motions once solutions are identified. In this study, we introduce a new approach called predictive reachability that decides reachability based on predicted trajectories. Our model utilizes a world model with a hierarchical policy. The world model allows the prediction of future trajectories and the evaluation of reachability. The hierarchical policy selects the embodiment based on the predicted reachability and plans accordingly. The proposed method does not necessitate explicit robot dynamics models but can perform learning based on partial observations. The evaluation results in basic reaching tasks across various environments demonstrate that our method outperforms previous model-based approaches in both sample efficiency and performance, while also enabling more reasonable embodiment selection based on predictive reachability.