Self-Learning Robot Manipulator Controller Using Reinforcement Learning

Traditionally, robot control has relied on mathematical solutions, particularly inverse kinematics (IK), to determine the joint angles needed for a desired end-effector pose. However, these solutions face significant challenges as robots increase in complexity with higher degrees-of-freedom (DoF) and adaptability requirements. This work introduces an approach for controlling robot manipulators by addressing the IK problem using the artificial intelligence (AI) technique. In this paper, a deep reinforcement learning approach is proposed to solve the IK problem for a 6-DoF industrial robot manipulator. The Deep Deterministic Policy Gradient (DDPG) algorithm is employed to train an agent capable of learning the robot's motion, allowing for random initial and target positions and orientations. Several reward functions are implemented to enhance the accuracy of the robot's movement towards the desired position and orientation. This proposed approach demonstrates a high success rate and low mean error, outperforming other state-of-the-art methods. Furthermore, the approach is validated on a physical robot, confirming its practical applicability and effectiveness.