GPU-based Grasp And Placement Planners For Sensor-modelled Objects

Abstract:
A Personal Robot should be able to handle unknown objects in unknown environments. For a manipulation task the question of what to do with an object once it has been grasped, is one of the most essential ones besides the grasping task itself. Moreover, the planning time should be at least as fast as the time the robot needs for its grasping/placing motions. We propose a combination of a fast grasp and placement planner that deals with sensor-modelled objects. They share novel pose computation steps that reduce the amount of computational intensive collision tests. By means of experiments we evaluate the computation time for the pose computation. Furthermore, we present the qualitative results of experiments showing accurate grasps and object placements.