Topological Synergies for Grasp Transfer

In this contribution, we propose a novel approach towards representing physically stable grasps which enables us to transfer grasps between different hand kinematics. We use a low dimensional topologically inspired coordinate representation which we call topological synergies, and which is motivated by the topological notion of winding numbers. We address the transfer problem as a stochastic optimization task and carry out motion planning in our topologically inspired coordinates using the Approximate Inference Control (AICO) framework. This perspective allows us to compute not only the final grasp itself, but also a trajectory in configuration space leading to it. We evaluate our approach using the simulation framework PhysX. The presented experiments, which develop further recent attempts to use topologically inspired coordinates in robotics, demonstrate that our approach makes it possible to transfer a large percentage of grasps between a simulated human hand and a 3-finger Schunk hand.