Vision-based Aided-Grasping in Teleoperation with Multiple Unknown Objects

摘要

Grasping an object in the scene with multiple unknown objects requires both the knowledge of which object to be the target and the planning of grasping pose. However, teleoperating a robot hand to find a proper grasping pose in an unstructured environment is often too complicated and time-consuming to perform. In this study, we propose an aided-grasping algorithm which autonomously corrects the pose of a robot hand using an eye-in-hand camera. We used multiple cameras for a natural vision-based teleoperation interface with an aided-grasping algorithm. Experiments with objects placed in arbitrary positions and angles have shown a successful implementation of the algorithm.