Grasping Points Discovery for Cloth Manipulation via Differentiable Physics-based Simulation

The selection of grasping points is essential in robotic manipulation tasks.  Especially when grasping deformable objects like clothes, unsuitable grasp points can lead to uncontrollable deformation. Previous works utilize learning-based approaches to estimate the manipulation performance based on human demonstration. However, clothes with a myriad of sizes and shapes have highly nonlinear structures, which makes the learning space forbidding large and discontinuous.  In this paper, we propose a novel grasping point optimization scheme that leverages gradient information from the differentiable simulation process as an estimate of performance differences between grasping point choices.  We derive the computation of gradients with respect to grasping points on the cloth surface in the DiffCloth differentiable cloth simulation. We conduct extensive experiments of clothes grasping tasks on manipulating hats, socks and bags and our gradient-based optimization scheme successfully optimizes grasp points to achieve optimal results.