Combined workspace monitoring and collision avoidance for mobile manipulators

Abstract

For safe human-robot interaction and co-existence, collision avoidance is a fundamental prerequisite. Therefore, in this contribution a Nonlinear Model Predictive Control approach for fixed-base and mobile manipulators is presented that allows for avoiding self-collisions and collisions with static and dynamic obstacles while performing tasks defined in the Cartesian space. The collision avoidance takes not only the end-effector but the complete robot consisting of both platform and manipulator into account and relies on a 3D obstacle representation obtained by fusing information from multiple depth sensors. The obstacle representation is applicable to all kinds of objects. It considers occlusions behind the obstacles and the robot to make a conservative assumption on the obstacle size. In order to achieve realtime reactions to obstacles, the obstacle information used in one control step is restricted to the most relevant obstacles determined by distance computation. The method is validated by means of simulation and by application to an omnidirectional mobile manipulator with 10 degrees of freedom.