A Complete Coverage Path Planning Algorithm for Mobile Robot Based on Improved Deep Reinforcement Learning

This article proposes an improved deep reinforcement learning algorithm for complete coverage path planning of mobile robots. The algorithm addresses the issue of agents easily getting trapped in local optima during training, which leads to long training times and high path redundancy. First, the working environment of the mobile robot is modeled, and the state dimensions are established. Then, a prioritized experience replay mechanism is introduced based on a double deep Q-network. Next, the action selection mechanism for the agent during training and testing is improved to prioritize effective actions, and when no effective actions are available, the A-star algorithm is invoked to quickly escape dead zones. Simulation results show that the proposed algorithm not only reduces the redundancy of coverage but also shortens the training time. Ablation experiment demonstrates that the stability of the proposed algorithm in training is superior to that of comparative algorithms, and it also validates the necessity of each component of the proposed algorithm.