Hybrid Robotic Grasping With a Soft Multimodal Gripper and a Deep Multistage Learning Scheme

Grasping has long been considered an important and practical task in robotic\nmanipulation. Yet achieving robust and efficient grasps of diverse objects is\nchallenging, since it involves gripper design, perception, control and\nlearning, etc. Recent learning-based approaches have shown excellent\nperformance in grasping a variety of novel objects. However, these methods\neither are typically limited to one single grasping mode, or else more end\neffectors are needed to grasp various objects. In addition, gripper design and\nlearning methods are commonly developed separately, which may not adequately\nexplore the ability of a multimodal gripper. In this paper, we present a deep\nreinforcement learning (DRL) framework to achieve multistage hybrid robotic\ngrasping with a new soft multimodal gripper. A soft gripper with three grasping\nmodes (i.e., enveloping, sucking, and enveloping_then_sucking) can both deal\nwith objects of different shapes and grasp more than one object simultaneously.\nWe propose a novel hybrid grasping method integrated with the multimodal\ngripper to optimize the number of grasping actions. We evaluate the DRL\nframework under different scenarios (i.e., with different ratios of objects of\ntwo grasp types). The proposed algorithm is shown to reduce the number of\ngrasping actions (i.e., enlarge the grasping efficiency, with maximum values of\n161% in simulations and 154% in real-world experiments) compared to single\ngrasping modes.\n