Contact localization through robot and object motion from point clouds

In this paper we address the problem of detecting contacts between a robot hand and an object during the approach and execution phases of manipulation tasks as a complement of touch perception. The motion produced on objects by unnoticed contacts is exploited to trigger the contact localization. Our method combines robot motion with tracking of target objects by means of point cloud to probabilistically estimate the location of contacts on a occupancy grid map. No previous knowledge on the object shape is required. The proposed approach is implemented on a real platform and experimentally validated in several cases.