Manipulation planning remains a difficult and largely unsolved problem in robotics. We consider the following version of the problem: one manipulator is available and the moving part can be grasped with a finite set of grasps. The planner has to plan automatically for any regrasping (if needed) to move the part from an initial configuration to a final configuration. We are working on PRM and investigate several issues in manipulation planning.
This animation (1.6MB MPEG) shows a PA10 robot reaching inside a narrow metal structure. The PA10, with 7 dof, is mounted on an invisible gantry, with additional 3 dof. The path was computed using a PRM and a simple geometrical path optimizer.
The goal is to use the PRM as a global path finder in a framework together with a local planer based on forward dynamics, to handle tasks like robot welding. The PRM finds a global feasible path from the initial position of the robot to the starting point of the weld seam. Once the weld seam is reached the forward dynamics planner takes over to handle the actual welding.
We are currently investigating the possibility of extending the PRM framework to handle manipulation planning. By using a method similar to the Lazy-PRM we have been able get some promising results.
This animation (4MB MPEG) shows the PA10 robot manipulating a large L-shaped object. To move the object from the initial to the goal position, the robot has to put down and re-grasp the object, which makes the task more difficult.