Inverse Kinematics of a 7-Degree-of-Freedom Robotic Arm Based on Deep Reinforcement Learning and Damped Least Squares

As we advance towards the future of the smart manufacturing industry, our research focuses on enhancing manipulator technology. Inverse kinematics is a key component of robotic arm control, yet many existing methods struggle to achieve high performance when dealing with high-precision target points and highly redundant robotic arms. In this paper, we propose a novel solution to the inverse kinematics problem by combining Proximal Policy Optimization (PPO) with the Damped Least Squares (DLS) method, forming the Multistep PPO-DLS Inverse Kinematics (MPDIK) algorithm. The algorithm was trained and tested in the PyBullet virtual environment, using random seven-dimensional position and pose target points. The MPDIK algorithm demonstrated outstanding performance, with the end effector achieving a distance error of less than 0.1 mm and an orientation error of less than 0.001°. Additionally, it exhibited excellent stability and fast convergence, with a post-training task completion success rate of 98.37% and an average of 20.68 time steps per task. This represents a significant improvement over existing methods, such as PPO and DLS, and demonstrates universal applicability. Our experiments also revealed that this method holds great potential for improving both the accuracy and real-time application capabilities of robotic systems.