Evolving polydisperse soft robotic jamming grippers

Abstract

Granular systems react to changes in external pressure by adapting their density through complex grain contact interactions. Granular packings subjected to small pressures are loose and fluidic, but are jammed into compressed, rigid packings at higher pressures. Common soft robotic jamming grippers are composed of a vacuum pump connected to a flexible membrane filled with granular material, e.g. ground coffee. The membrane encompasses an object, and the pump activates. The grains jam, deforming the membrane, and grasping the object, with grain morphology playing a critical role in determining gripper performance. Bespoke grippers can be designed to effectively grasp specific objects by evolving their constituent grains. Evolved grippers have to-date used exclusively monodisperse granular materials (grains with identical size and shape). However, while not conceived through evolution, polydisperse grippers comprised of natural grains varying in size and morphology can often perform better in real-world grasping experiments. We employ the Discrete Element Method and NSGA-III to optimise grasps on disparate objects by evolving distributions of superellipsoidal grains (varying both their shapes and volumes) within the gripper. Results elucidate the successful application of multi-objective evolution to design bespoke polydisperse jamming grippers, and how variations in grain surface curvatures and volumes influences grasping performance.