Determining Grasp Positions with 4-Finger Gripper Manipulator Using Reinforcement Learning

This paper discusses the use of a 4-finger gripper manipulator and a reinforcement learning-based approach to enable a robot to grasp a wide range of objects with different shapes and sizes. We especially focused on proper manipulator posture, as well as various components of reinforcement learning-based approaches, including learning environment structures and reward forms. The reward function was set to take into account the three-dimensional Euclidean distance between the gripper’s point and the target, the direction of the gripper, and the difference in the target’s orientation. In this study, we used gazebo simulation and Robo-gym framework and had good experimental results.