Abstract: In order to solve the limitations of the existing autonomous navigation technology of coal mine robot, a visual navigation technology based on RGBD in coal mine scene is proposed. The overall architecture of visual navigation is studied, and each functional module is designed and described; the map creation method is studied, the pose relationship between consecutive frames generated by ORB_SLAM2 algorithm is transformed, and the dense point cloud map is obtained by frame RGBD point cloud splicing; the implementation process of autonomous positioning and path planning of inspection robot is designed. Combined with Monte Carlo, D* and DWA algorithms, the accurate positioning and navigation of robot are completed; the test platform of the visual navigation technology is established, and a series of simulation tests are carried out using the platform. The test results show that the visual navigation technology based on RGBD in the coal mine scene can meet the autonomous walking function requirements of the wheeled inspection robot in the coal mine, with good positioning accuracy.