Research on Robot Path Planning Based on Deep Reinforcement Learning with Big Data

Path planning is a core technology for mobile robots to achieve autonomous navigation and intelligent operations. Against the backdrop of the development of intelligent robotics, traditional path planning algorithms, such as A* algorithm, artificial potential field method, and bionic algorithms, have limitations in handling complex environments. Reinforcement learning, especially Q-learning and its improved algorithms, offers new solutions for path planning. However, traditional Q-learning faces challenges like the curse of dimensionality, and DQN has the problem of overestimation. This paper proposes a robot path planning method based on Deep Double Q-Learning (DDQN). It constructs a Q-value function approximation model using a deep neural network, avoiding the curse of dimensionality through large-scale data-driven end-to-end training. The DDQN algorithm is adopted to alleviate the overestimation issue, and a re-replay mechanism is proposed to improve convergence. Additionally, the Bézier curve is used to smooth the planned path. Simulation experiments in a 40×40 grid map environment show that compared with the traditional Q-learning algorithm, the proposed algorithm shortens the average path length by 32.5%, reduces the number of turning points by 42.9%, decreases the average curvature fluctuation value, and increases the convergence speed by 57.1%, demonstrating higher adaptability and efficiency in robot path planning.