A Motion Planning Framework for Skill Coordination and Learning

Coordinated whole-body motions are key for achieving the full potential of humanoids performing tasks in human environments and in cooperation with humans. We present a multi-skill motion planning and learning framework which is able to address several complex whole-body coordinations and as well detection and imitation of motion constraints. The framework is composed of three main parts: first, a minimal set of basic motion skills is defined in order to achieve basic stepping, balance and reaching capabilities. A multi-skill motion planner is then developed for coordinating motion skills in order to solve generic mobile manipulation tasks. Finally, learning strategies are presented for improving skill execution and for learning constraints from imitation. The framework is able to coordinate basic skills in order to solve complex whole-body humanoid tasks and also integrates learning mechanisms for achieving humanlike performances in realistic environments.