Combining Task and Path Planning for a Humanoid Two-arm Robotic System

The ability to perform both causal (means-end) and geometric reasoning is important in order to achieve autonomy for advanced robotic systems. In this paper, we describe work in progress on planning for a humanoid two-arm robotic system where task and path planning capabilities have been integrated into a coherent planning framework. We address a number of challenges of integrating combined task and path planning with the complete robotic system, in particular concerning perception and execution. Geometric backtracking is considered: this is the process of revisiting geometric choices (grasps, positions etc.) in previous actions in order to be able to satisfy the geometric preconditions of the action presently under consideration of the planner. We argue that geometric backtracking is required for resolution completeness. Our approach is demonstrated on a real robotic platform, Justin at DLR, and in a simulation of the same robot. In the latter, we consider the consequences of geometric backtracking.