Combining task and path planning for a humanoid two-arm robotic system

Abstract

The ability to perform both causal (means-end) and ge-
\nometric reasoning is important in order to achieve au-
\ntonomy for advanced robotic systems. In this paper, we
\ndescribe work in progress on planning for a humanoid
\ntwo-arm robotic system where task and path planning
\ncapabilities have been integrated into a coherent plan-
\nning framework. We address a number of challenges
\nof integrating combined task and path planning with
\nthe complete robotic system, in particular concerning
\nperception and execution. Geometric backtracking is
\nconsidered: this is the process of revisiting geometric
\nchoices (grasps, positions etc.) in previous actions in
\norder to be able to satisfy the geometric preconditions
\nof the action presently under consideration of the plan-
\nner. We argue that geometric backtracking is required
\nfor resolution completeness. Our approach is demon-
\nstrated on a real robotic platform, Justin at DLR, and
\nin a simulation of the same robot. In the latter, we con-
\nsider the consequences of geometric backtracking