Abstract: With the continuous promotion of intelligent construction in coal mines in China, the development of mine auxiliary transportation vehicles towards unmanned driving has become an inevitable trend. As the core unit of unmanned vehicles, the positioning system cannot meet the full process, high-precision, and low time delay positioning requirements of coal mine auxiliary transportation vehicles from the ground fabric field, inclined shaft roadway, underground roadway, to mining face with a single positioning method and traditional positioning algorithms. Firstly, based on the operating conditions of coal mine auxiliary transportation vehicles and the tunnel environment, a seamless positioning system for underground and above mines based on GNSS/UWB fusion IMU was designed, and a model switching delay (D_MS) was proposed as the performance evaluation index for the seamless positioning system; Secondly, to address the issue of non line of sight (NLOS) errors in UWB positioning, a UWB/IMU tight combination downhole positioning algorithm was designed, and Error State Kalman Filter (ESKF) was used to filter and optimize it. Simulation results showed that the ESKF optimization algorithm had an average positioning error of 0.19 m, with an accuracy improvement of 56% compared to single UWB positioning; Once again, the influencing factors of interactive multiple models were analyzed. In response to the problem of large model probability matrix errors affecting seamless positioning accuracy, a mine seamless positioning algorithm based on ESKF and fuzzy adaptive improved interactive multiple models (FAIMM-ESKF) was designed. Simulation results showed that the positioning accuracy of the FAIMM-ESKF algorithm was improved by 29% compared to before improvement; Finally, a simulated inclined shaft tunnel was constructed in the laboratory, and a seamless positioning system positioning and evaluation experiment was conducted using an unmanned test vehicle. The results showed that the average error of the seamless positioning system in the interaction area between the well and the underground was 0.131 m, and the maximum error was 0.452 m, which was reduced by 17.6% and 14.8% compared to traditional algorithms, respectively; Throughout the entire experimental process, the maximum error of the FAIMM-ESKF algorithm was 0.498 m, the average error was 0.25 m, and the average model switching delay was 35 ms, which can meet the positioning accuracy and delay requirements of unmanned driving in the entire process of coal mine auxiliary transportation vehicles. The research results can provide theoretical reference for promoting the establishment of a seamless connection, precise and efficient positioning system and positioning algorithm for coal mines, and have important theoretical significance and practical value for accelerating the normalization of unmanned driving of auxiliary transportation vehicles in coal mines and accelerating the intelligent construction of coal mines.