In order to realize automatic vehicle navigation in agricultural facilities without GPS (global position system) signal, a path tracking method based on UWB (ultra wideband) positioning was proposed in this study. The test prototype uses a pure electric and four wheels mechanism with 2SW-2DW (2 steering wheels and 2 drive wheels) structure. The closed-loop system with the controller, driver, motor and encoder ensures the accuracy of the control. The indoor UWB wirelesspositioning system based on the TOA (time of arrival) principle was built by using 4 anchor nodes, and a WLS (weighted least squares) method was used to solve the statically indeterminate equations, which improved the positioning accuracy of mobile tags. The positioning error of the UWB positioning system in the 16 m×11 m rectangular center area is within 7 cm, and the positioning error of the rectangular edge is less than 12 cm, which meets the positioning requirements of path tracking test in agricultural facilities. In view of the limitation that the lookahead distance must be greater than the lateral deviation in the traditional pure pursuit model, the lookahead distance was redefined in this paper. So the scope of application of the pure pursuit model had been widened. The deviation degree of vehicle was defined quantitatively according to the angle between the heading of the vehicle and the lookahead line. The traditional pure pursuit model was improved, and a new pure pursuit model algorithm based on dynamic lookahead distance was proposed to further improve the quality of path tracking. The algorithm of path tracking was simulated and verified using MATLAB 2016a. The results showed that the average error, the maximum deviation and the stability distance of the improved algorithm with dynamic lookahead distance were better than those of the traditional pure pursuit algorithm with fixed lookahead distance, indicating that the proposed improved algorithm is effective theoretically. The real vehicle test results showed that in the guidance of UWB positioning system, the vehicle could converge to the desired path in different initial states. Linear tracking were carried out in 4 initial states. When the test prototype speed is 0.5 m/s and the signal sampling period is 10 Hz, the average deviation, maximum deviation, stable distance and adjustment time are 23.3-35.0 cm, 39.0-124.6 cm, 80.3-283.0 cm, and 6.6-13.5 s, respectively. It was also observed that the above corresponding index increased with the initial deviation of the prototype. When the prototype reaches a stable state, the steady-state deviation is 5.4-8.4 cm, and its average steady-state deviation is 6.3 cm. In the rectangular path tracking, the overall mean deviation is 20.6 cm when the initial lateral deviation and heading deviation are 0. The lateral deviation mainly occurs at 90° turning, and the maximum deviation is 85.5 cm. The path tracking quality of the improved pure tracking algorithm with dynamic lookahead distance is better than the traditional pure tracking model with fixed lookahead distance, and it can meet the requirements of mobile platforms in frequent steering for automatic navigation in the agricultural facilities. This method could provide a new idea for vehicle navigation in agricultural facilities.