Path Planning for Robot Inspection Coverage Based on Deep Reinforcement Learning

Robotic inspection serves as a critical monitoring method for construction sites, railway corridors, and large-scale facilities, where deployment strategies directly impact inspection coverage efficiency. To address the limitations of traditional planning methods—which rely heavily on human experience and heuristic algorithms and struggle to achieve global optimality in complex environments—we propose a patrol coverage strategy for mobile robots based on the Policy-Based Proximal Optimization (PPO) algorithm. By utilizing deep reinforcement learning to establish an interaction mechanism between the agent and the simulation environment, the robot deployment process is modeled as a Markov Decision Process (MDP). Regarding the design of the reward function, this paper introduces dynamic scarcity gains and coverage redundancy penalties in addition to conventional factors such as coverage increment, deployment cost, and collision penalties. Through this refined reward mechanism, the optimal deployment of inspection locations is achieved. Subsequently, by combining the Asymmetric Traveling Salesman Problem (ATSP) with the A* algorithm, the task is decoupled into two sequential stages—layout optimization and path planning—to generate optimal inspection trajectories. Experimental results show that the method achieves an average coverage rate of 96.47%, a 5.56% improvement over the baseline algorithm, and an average of 37.6 deployment points, a reduction of 14.6 points compared to the baseline. It successfully generates optimal inspection paths across three random experimental scenarios. Through the deep integration of deep reinforcement learning and path planning, this study provides an efficient, smooth, and highly generalizable decision-making framework for autonomous coverage inspection by mobile robots in complex geometric environments.