Model Predictive Interaction Control based on a Path-Following Formulation

This paper presents an approach to predictive path-following control for robotic manipulation tasks. In addition to free motions, interactions of the end effector with environmental objects are also taken into account. This requires a variable control structure including motion control, force control, and hybrid combinations, which is provided by the model predictive interaction control (MPIC) framework. In the presented approach, MPIC is reformulated as a path-following control and used within a hierarchical framework for the systematic description of manipulation tasks. In order to obtain a meaningful physical interpretation of the path progress parameter, an approach to normalize the individual control errors for joint angles, Cartesian pose as well as interaction forces is proposed. This allows to reinterpret the path parameter as progress along the overall task. Finally, the approach is experimentally validated on a 7-degree-of-freedom (DOF) industrial robot and a realtime capable control cycle of 10 ms.