Stochastic Force-Closure Grasp Synthesis for Unknown Objects Using Proximity Perception

Abstract

Proximity perception is a promising technology that provides near-field information valuable for robotics. To improve the functionality of proximity perception in anthropomorphic hands, we propose a novel stochastic force-closure grasp synthesis (SFCGS) that finds robust grasps insensitive to object uncertainty introduced by the lack of explorations and perception noise. Specifically, we propose a dual-mode perception system comprised of five flexible capacitive dual-mode (proximity and pressure) sensors. Using the Gaussian process, we explore unknown objects with proximity perception and approximate their signed distance function. The SFCGS formulates the problem of finding the probabilistically optimal grasps as minimizing the separation probability of the origin and stochastic grasp wrench space. In addition, a dual-mode reactive controller is presented to improve the grasping success rate. The results from simulation experiments indicate that proximity perception can reduce the reconstruction error by 4.53 cm compared to tactile perception. Furthermore, the newly introduced SFCGS can yield more uncertainty-insensitive grasps than the traditional force-closure approach. In real-world experiments, the proposed approach achieves a considerable 13.4% improvement in success rate over benchmark methods. The outcomes of this study are significant in promoting the application of proximity perception in robot hand-arm systems and upper-limb prostheses.