Exploring the Application of Particle Filters to Grasping Acquisition with Visual and Tactile Occlusion

The G-SL(AM)2 (or grasp-SLAM) problem is the analog of the well studied Simultaneous Localization and Mapping (SLAM) problem in the domain of grasping. The G-SL(AM)2 problem is to localize the object relative to the hand while simultaneously updating the system model and manipulating the object. Stick-slip friction and intermittent contact are fundamental aspects of grasping that lead to nonsmoothness of the mathematical model. For this reason, a particle filter is designed and its effectiveness is explored through a simplified instance of the G-SL(AM)2 problem. Experiments from a planar grasping testbed are used to test the ability of the particle filter to track the object and adjust friction model parameters with or without visual occlusion and with or without binary tactile sensor information. The results show that the particle filter outperforms a simple Kalman filter especially during periods of visual occlusion.