The Grasping Force Control for Force Sensor-less Robot through Point Clouds Mask Segmentation

It's difficult for a robot to adjust its sensor-less parallel gripper to a proper degree, holding the object with a tolerable deformation and no slipping, when operating deformable objects with unknown physical property (OUPP). This paper introduces a grasping strategy to solve this problem. Firstly, a semantic segmentation model was applied to monitor the deformation of the object by generating the cloud points of the region. And this deformation was treated as a criterion to measure whether the grasping force overload or not. Based on the strategy, a cup grasping experiment was conducted, which turned out a 90 % success rate with the proportion of deformation less than 2%. Thus, it demonstrates the strategy proposed can improve the performance of robots with sensor-less end-effector on OUPP.