关键词: 移动机器人, 路径规划, 深度Q网络算法, 蚁群算法, 强化学习, 算法优化

Abstract: To address the issues of slow convergence and poor path planning associated with the Deep Q-Network (DQN) algorithm for mobile robot path planning in large-scale complex unknown environments, a path planning algorithm combining Ant Colony Optimization (ACO) and DQN, termed ACOG-DQN, is proposed. Initially, the pheromone mechanism of ACO is introduced to facilitate the selection of potential paths with the goal of reaching the destination, thereby reducing the number of ineffective environmental explorations and determining the optimal path. Concurrently, the previous path selection experiences are filtered using a threshold to form a sample set for training the Q-network, which is then utilized to determine the optimal path for the mobile robot in the current environment. Finally, a path selection mechanism is designed where the weight of the Q-network's optimal path increases over time, using the optimal paths determined by ACO and the Q-network, as well as those determined by random exploration, as candidates. This mechanism selects the current action, aiming to achieve a path that is ultimately decided entirely by the Q-network. Simulation and physical experiments conducted in three different complex environments demonstrate that the proposed ACOG-DQN algorithm exhibits superior performance in terms of convergence speed, path quality, and algorithm stability compared to the DQN algorithm, thereby validating the effectiveness of the proposed algorithm.

Key words: mobile robot, path planning, Deep Q-network(DQN), ant colony optimization algorithm(ACO), reinforcement learning, algorithm optimization