EfficientGrasp: A Unified Data-Efficient Learning to Grasp Method for Multi-Fingered Robot Hands

Autonomous grasping of novel objects that are previously unseen to a robot is\nan ongoing challenge in robotic manipulation. In the last decades, many\napproaches have been presented to address this problem for specific robot\nhands. The UniGrasp framework, introduced recently, has the ability to\ngeneralize to different types of robotic grippers; however, this method does\nnot work on grippers with closed-loop constraints and is data-inefficient when\napplied to robot hands with multigrasp configurations. In this paper, we\npresent EfficientGrasp, a generalized grasp synthesis and gripper control\nmethod that is independent of gripper model specifications. EfficientGrasp\nutilizes a gripper workspace feature rather than UniGrasp's gripper attribute\ninputs. This reduces memory use by 81.7% during training and makes it possible\nto generalize to more types of grippers, such as grippers with closed-loop\nconstraints. The effectiveness of EfficientGrasp is evaluated by conducting\nobject grasping experiments both in simulation and real-world; results show\nthat the proposed method also outperforms UniGrasp when considering only\ngrippers without closed-loop constraints. In these cases, EfficientGrasp shows\n9.85% higher accuracy in generating contact points and 3.10% higher grasping\nsuccess rate in simulation. The real-world experiments are conducted with a\ngripper with closed-loop constraints, which UniGrasp fails to handle while\nEfficientGrasp achieves a success rate of 83.3%. The main causes of grasping\nfailures of the proposed method are analyzed, highlighting ways of enhancing\ngrasp performance.\n