A fast heuristic Cartesian space motion planning algorithm for many-DoF robotic manipulators in dynamic environments

A variety of motion planning algorithms has been developed over the past decades. For robotics, the transformation of the problem from the robot Cartesian space (workspace) to the configuration space (C-space) has been crucial - converting the problem of collision-checking between 3D objects to simpler, point-like, tests in the high-dimensional C-space. However, the C-space map-making is both computationally and memory-intensive and the complexity grows with the C-space dimension. With time-varying environments and many robot DoFs, this soon becomes intractable, as newly appearing and possibly moving obstacles need to be remapped online into the high-dimensional C-space. Therefore, we present a fast heuristic planning method designed for a humanoid robot that employs the sampling-based RRT* algorithm directly in the Cartesian space and in a hierarchical fashion: first, a collision-free path is planned for the end-effector; second, corresponding collision-free points for every via-point are searched for the robot elbow. The resulting path consists of straight-line segments and is only approximate, but the details - a kinematically feasible smooth trajectory for the robot - is offloaded to an online Cartesian solver and controller that is available on our platform. The results demonstrate, first, that our solution delivers real-time performance (plans faster than 1s on a standard PC) in the vast majority of cases in a significantly cluttered environment. Second, the results are suggestive of the fact that asymptotic optimality of the plans is preserved even for the additional control points. Third, a test of state-of-the-art algorithms on the same scenario shows that solutions cannot be found in reasonable time (less than 10s).