Active Visuo-Haptic Object Shape Completion

Recent advancements in object shape completion have enabled impressive reconstructions using only visual input. However, due to self-occlusion, the high uncertainty occluded parts, which negatively impacts performance of downstream robotic tasks such as grasping. In this work, we propose an active visuo-haptic method called Act-VH that actively computes where touch objects based on reconstruction uncertainty. reconstructs from point clouds and calculates IGR, a recent state-of-the-art implicit surface deep neural network. We experimentally evaluate accuracy against five baselines simulation real world. also new environment for purpose. The results show outperforms all uncertainty-driven haptic exploration policy leads higher than random driven by Gaussian Process Implicit Surfaces. As final experiment, best baseline grasping 10 novel objects. reaches significantly grasp success rate Together, work opens up door more complex cluttered scenes.