SoGraB: A Visual Method for Soft Grasping Benchmarking and Evaluation

摘要

The need for industrially relevant tools to safely handle delicate and deformable goods has driven a recent explosion in soft robotic gripper designs. However, there is currently no meaningful way to compare different designs. No commonly available, standardised evaluation protocol exists to assess the performance of soft grippers. This work introduces the Soft Grasping Benchmarking and Evaluation (SoGraB) method to evaluate grasp quality. It quantifies object deformation, and hence grasp quality, by measuring the Density-Aware Chamfer Distance (DCD) between point clouds of soft objects recorded before and after grasping. Through extensive experimentation, we demonstrate the method’s ability to evaluate the quality of soft grasps, rank different gripper designs, select soft grippers for complex grasping tasks, and benchmark them for comparison against future designs. We believe SoGraB can be a standard for grasp evaluation and invite future users to contribute by benchmarking their own soft designs against our baselines or adding objects to the dataset.