Collision-Free Motion Generation in Dynamic Environments with Joint-Space Quadratic Optimization

Abstract

Service robots acting as human assistants are required to have the ability of adjusting their movements to accommodate changes in the interaction. In this paper, a collision-free motion generation framework of manipulators is proposed for physical interaction tasks. First, virtual ellipsoids are introduced to remove the redundant path points generated by the rapidly-exploring random tree algorithm. The motion margin of the end-effector can be ensured at the same time. Second, a reactive replanning strategy is proposed that allows the manipulator to perceive potential contact. A safe path for avoidance is found before further collisions occur. Finally, a joint-space trajectory generator based on quadratic programming is formulated to produce continuous dynamic motions under multiple hardware limits. Experiments demonstrate that the proposed method enables the manipulator to effectively avoid dynamic obstacles online, and that the motion of all joints is within the allowed range.