In-hand object localization has always been a critical but difficult aspect of dexterous robotic manipulation. We attempt to address this issue in this paper through the use of point cloud registration techniques. Specifically, the grasping pose is estimated by registering the high-resolution 3D contact point cloud sensed by a novel GelStereo tactile sensor with the object template point cloud. Extensive qualitative and quantita- tive analyses of in-hand localization and insertion experiments of small parts are performed on our robot platform. The experimental results verify the accuracy and robustness of the proposed in-hand object localization pipeline.
