A Path Planning Approach Based on Q-learning for Robot Arm

A path planning method based on Q-learning is proposed for robot arm. As reinforcement learning, Q-learning is widely used in the field of mobile robot navigation due to its simple and well-developed theory. However, most of researchers avoid using it for solving the robot arm path planning problem because it has to take each joints motion into account. In this study, approximate regions instead of accurate measurements are used to define new state space and joint actions. Moreover, the reward function takes the distance from current position of robots' end-effector to the goal position into account. The experimental simulation shows that the Q-learning approach is efficient and has ability to plan a collision free path for robot arm when the order of magnitude of state space is below 10
⁴
. Furthermore, the experimental results indicate that the number of obstacles affects the calculation time for the same iterations. The more obstacles, the less calculation time the algorithm needs. The weight coefficient in the reward function affects the convergence speed and the quality of the solutions.