Path planning for redundant manipulator without explicit inverse kinematics solution

Path planning for redundant robotic manipulators received continuous interest in the past decades. Most efforts focused on random sampling-based methods, such as probabilistic roadmap method (PRM) and rapidly-exploring random tree (RRT), since they are suitable for planning in high-dimensional configuration space. Given the workspace goal position and orientation of the end-effector, however, explicitly calculating a collision-free and reachable goal configuration for robot joint angles in the presence of joint limits and self-collisions is not a trivial work. The difficulty forms a bottleneck for the broader applicability of the randomized path planning methods. In this paper, a novel two-stage approach is presented to implicitly solve the formation of inverse kinematics (IK) problems, which employs a variant of RRT to embed the process of IK calculation into construction and exploration of the tree-based data structure. Combined with bidirectional RRT-connect algorithm, the two-stage approach can efficiently address the path planning problem for general redundant manipulators. The algorithm has been implemented and several 2D and 3D experiments demonstrate the effectiveness of the method.