Continual Learning for Anthropomorphic Hand Grasping

Wanyi Li, Wei Wei, Peng Wang

Abstract: It is important and challenging to learn to grasp different objects with anthropomorphic robotic hands continually and incrementally. However, most current works do not have this property. They learn grasp planners using large preprepared datasets, do not generalize well to new objects and are difficult to improve continually. Besides, existing continual leaning works rarely target at anthropomorphic hand grasping, and usually deal with short streams of experiences. Because of the intrinsic long stream nature of anthropomorphic hand grasping, it is hard to utilize off-the-shelf continual learning (CL) methods for it. In this article, we propose to introduce continual machine learning into anthropomorphic hand grasping and design the CL framework of anthropomorphic grasping (CLFAG framework). It includes three modules: 1) Data Producer; 2) Grasp Experiences; and 3) CL Algorithm $A_{\mathrm{ CL}}$ , thus, makes the CL of anthropomorphic grasping possible. To overcome the catastrophic forgetting problem in long streams of grasping experiences, we propose a CL algorithm based on importance-based regularization and diversity-aware replay within the CLFAG framework. Furthermore, we construct a dataset for CL of anthropomorphic grasping. Experiments on constructed dataset and in simulation demonstrate the effectiveness and superiority of the proposed approach.