Toward dynamic manipulation for humanoid robots: Experiments and design aspects

This paper discusses a control architecture to equip humanoid robots with dynamic manipulation skills. Such skills are beneficial, as they increase the dexterity of humanoid robots and improve their handling of unforeseen situations. The nature of manipulation poses challenges with respect to environment perception, action planning, and motion control. These challenges are addressed in the paper: a dynamic force/torque (F/T) observer is discussed to reconstruct the environment F/Ts and to improve the interaction control. A method for online motion planning is investigated, which generates trajectories based on different selection criteria. These elements are integrated in a control design for dynamic manipulation and experimentally validated for two tasks: two-handed ball throwing and one-handed ball catching. Based on the experimental results, a modified end effector design with intrinsic compliance is proposed to improve the system performance that is demonstrated in case studies.