Learning based regrasp policy from tactile feedback

Abstract

In the context of robotic object manipulation, this work presents a simple regrasp policy based on the tactile feedback captured by the "fingers" of the robot gripper. To do so, there is a learning based function that assesses the quality of a grasp and another model based methodology that searches for better grasping points. These algorithms have been tested on a wide variety of unknown objects, obtaining a significant grasp success improvement.

Outgoing