Abstract: Traditional coal mine inspection robot obstacle avoidance methods often rely on preset maps and simple sensor feedback, making it difficult to cope with the complex dynamic environment of coal mines, leading to failure or low efficiency in obstacle avoidance. Therefore, research has been conducted on an SLAM-based autonomous obstacle avoidance method for coal mine inspection robots. First, a stereo vision system is used to acquire environmental information, enabling obstacle detection. By optimizing the SLAM algorithm using graphs, the optimal pose estimation results for the inspection robot are obtained. Combining the results of obstacle detection and pose estimation, the A* algorithm is employed to plan an autonomous obstacle avoidance path, achieving both global path planning and local dynamic obstacle avoidance. Experimental results show that under conditions of complex spatial distribution of obstacles and dynamic interference, the coal mine inspection robot can continuously and stably advance towards the target point, successfully navigating around obstacles without collision. Compared to traditional methods, this approach offers better obstacle avoidance paths and shorter avoidance times, demonstrating excellent stability and scene generalization capabilities in dynamic and complex environments.