Grasp Moduli Spaces, Gaussian Processes and Multimodal Sensor Data

Poster (konferens), 2014

We discuss our recent and ongoing work on deformation-based grasp synthesis and transfer which formulates a joint space of shapes and grasps – a Grasp Moduli Space– within which both grasp configurations and object shapes can be continuously deformed. In this context, we propose the use of Gaussian Process-based implicit surface representations. These shape representations complement our previous work on using spherical harmonics as well as explicit cylindrical coordinates to model the object shape component of a Grasp Moduli Space. We provide a preliminary experiment with these Gaussian Processes showing how shape representations can be obtained using multimodal visual and haptic information and discuss how these representations can be continuously deformed for the purpose of transferring and generalizing known grasps.