Abstract: In order to meet the requirements of high-precision mapping and safe navigation for robot inspection in complex underground coal mine environment, and solve the problems of insufficient adaptability, anti-interference and path planning coordination of traditional algorithms in underground coal mine, the research on coal mine robot mapping and navigation based on improved lidar odometry and mapping (LOAM) algorithm was carried out. In the construction of environmental map model, based on simultaneous localization and mapping (SLAM) framework, the point cloud data collected by lidar is converted into grid map. Aiming at the dust interference in coal mine, the dynamic filtering method based on Gaussian probability model is introduced. The low probability invalid points are eliminated by calculating the effective probability of measurement points, and the threshold is automatically adjusted according to the statistical characteristics of dust concentration. In the process of feature point extraction, geometric features and texture features are fused, and the sub dimension of feature description is supplemented by calculating the gray change gradient in the neighborhood of point cloud, so as to improve the accuracy of feature point matching and the quality of map construction. In terms of path planning, based on the improved A* algorithm, a dynamic weight mechanism is adopted to dynamically adjust the weights of the evaluation function according to the density of environmental obstacles, thereby optimizing the global path planning. The dynamic window approach (DWA) is used for local path planning, and the trajectory curvature penalty item is added to suppress the rapid turning motion of the robot, and the two are integrated to realize hierarchical navigation. Simulation results show that the absolute error of map construction based on improved LOAM algorithm is controlled within the range of −0.053 m to 0.035 m, and the relative error is less than or equal to 2.65%. In different dust concentration environment conditions, both the running time and the maximum relative error are better than those of the comparison algorithm. Compared with the improved bat algorithm and the improved fast expansion random tree algorithm, the path planning fusion algorithm has a running time of about 2 s, a shorter path length, and a maximum obstacle avoidance success rate of 98.8%. In the real coal mine environment test, a 500 m roadway map was constructed in a 2 000 m undergrand roadway in a coal mine in Shanxi after continuous operation for 8 h. The absolute error range is narrowed down to −0.042 m to 0.028 m, the maximum relative error was 3.2%, the average running time of navigation algorithm was 1.8 s, and the success rate of obstacle avoidance was 96.7%. The improved LOAM algorithm and path planning fusion algorithm can improve the mapping accuracy and navigation efficiency of coal mine robot in complex underground environment.