Simultaneous Tactile Exploration and Grasp Refinement

Other conference contribution, 2022

In this work, we present a method for simultaneous exploration and grasping of unknown objects by combining tactile and visual information. In many robotic applications, visual data is incomplete due to occlusions. We show how Gaussian processes implicit surfaces (GPIS) can be used to model an initial representation of shape from visual information. This perceived partial model can then be augmented from tactile glances obtained during grasp attempts. Finally, grasp planning is done by means of a bi-objective optimisation method, in which we simultaneously optimise the grasp probability of success and improve the object’s shape representation during the execution of successive grasps. Experimental results in simulation demonstrate the efficiency of our approach.