Abstract: The driving condition of underground unmanned vehicle in coal mine is complicated. There are continuous downhill when the vehicle goes down the mine, and continuous uphill; The underground road surface is slippery and uneven, and the road adhesion coefficient is low. There are many forks in underground roadway, and the turning angle is large. There are many problems such as high power equipment and serious interference of strong magnetic field. The existing distributed drive autonomous electric vehicle stability control method is difficult to achieve vehicle stability and safety under multiple working conditions under complex conditions in coal mines, and there is no GPS signal in coal mines. A state estimation method of unmanned vehicle based on LiDAR and IMU is proposed. By using IMU to correct the distortion of LIDAR data, the noise generated by LIDAR in the process of moving is reduced. The adaptive module and time series analysis are introduced into UKF to improve the estimation accuracy of the status of unmanned vehicles in coal mines.According to the state parameters of unmanned vehicles in underground coal mine, the typical operating conditions of unmanned vehicles in underground coal mine are classified, and different state equations are designed. A neural network multi-condition matching degree analysis method is proposed, which can output the matching degree between the current working condition and the typical working condition, and make a weighted fusion with the output control quantity of MPC, so as to ensure the optimal stability of the unmanned vehicle under different working conditions and the smoothness of the vehicle when the working condition is changed. The boundary between different working conditions is weakened, and the generalization of MPC under different underground working conditions is improved. Simulation experiments and real vehicle experiments were carried out for the state estimation method and stability control method of unmanned vehicles in coal mines. The experimental results showed that compared with conventional MPC controllers, the performance of multi-model MPC controller was improved, and the overall system was effective under actual operating conditions in coal mines.