Abstract: Simultaneous localization and mapping (SLAM) is a key technology to achieve coal mine intellectualization, but its usage is restricted in underground coal mines due to feature degradation. To address this problem, a multi-sensor fusion SLAM algorithm which is suitable for underground coal mines is proposed. It consists of two sub-systems: a visual odometry system and a lidar SLAM system. The visual odometry system is composed of a near-infrared camera and an inertial measurement unit (IMU) sensor. The laser SLAM system is based on the features-point laser SLAM framework, using visual odometer information instead of IMU pre-integration, and improving the laser point cloud feature classification method for coal mine roadway structure to optimize the radar frame scanning matching. The exception handling mechanism is designed in the vision odometer system to avoid IMU error accumulation caused by point cloud feature degradation, which leads to the failure of positioning and mapping. The test results of the algorithm in simulated roadway of the coal mine show that the algorithm can run reliably in the roadway environment, and the stability and robustness of the algorithm are significantly improved compared with the existing SLAM algorithm.