Training an Under-actuated Gripper for Grasping Shallow Objects Using Reinforcement Learning

Robot programming and training depends on the task that needs to be completed, the end-effector properties and functionalities and the working space. These considerations can complicate the programming process, which in return, increase the time that is needed for training the robot. Thus, several research approaches have been introduced to address training the robots intuitively. In this regard, this paper presents an approach for training an under-actuated gripper and the robot attached to it for grasping shallow objects. The research work started by detailed analysis of the fingers of human hand during the grasping process. Then, a modified design of the gripper has been produced. This modification includes adding an artificial nail among other hardware-related modifications. Then, a Q-Learning algorithm has been used for training the gripper on grasping the shallow object. With two fingers, three actions were configured, and 625 states were configured for the learning algorithm. For the validation, a coin has been used for representing the shallow object. The results showed reduction in both the grasping time and the number of movements.