Towards a Generic Manipulation Framework for Robots based on Model Predictive Interaction Control

Abstract

This paper presents an optimization-based control framework for robotic manipulation tasks. A hierarchical and systematic decomposition is used in order to formulate the high-level task as a sequence of manipulation primitives. A key difficulty is that different primitives usually require different strategies to control motions and forces. However, based on the concept of model predictive interaction control (MPIC), it is possible to use the same control approach for all primitives. Instead of changing the controller, only the parameterization of the MPIC, i.e. the cost function and the constraints, is adapted for each primitive. The hierarchical task decomposition and the generic control concept are demonstrated for a screwing application using a robot arm with seven degrees of freedom.