Abstract: The current control accuracy of automatic navigation curve tracking in tractors cannot fully meet the production needs in modern agriculture. In this study, a tracking control system of variable curvature path was presented as the front wheel angle feedforward compensation. Firstly, a comprehensive evaluation was made on the influence of agricultural machinery operating speed and reference path curvature on the foresight distance. The lookahead area was then altered to compute the preview point, in order to realize the dynamic adjustment of the lookahead distance and the feedforward amount of the front wheel angle. Furthermore, the lookahead distance posed a significant effect on the speed of the vehicle. Specifically, the vehicle often traveled at a lower speed on the rugged and curved reference path, when the tractor was working on the field mechanized road or irregular fields. However, the lookahead distance was typically expressed as a linear function that related to the speed of the vehicle. The inaccurate tracking of the reference path was normally obtained due to the weak coupling, if only the speed of the vehicle without considering the curvature of the driving road on the lookahead distance. Therefore, the varying lookahead distance was proposed to depend on the curvature of the path. Meanwhile, the varying curve length of the preview point in the lookahead area was relative to the starting point of the lookahead area depending on the curvature of the reference path. Secondly, the steady-state error was often observed in the driving process of the agricultural machinery, because the pure pursuit algorithm failed to consider the deviation between the current pose and the reference path of the agricultural machinery, when tracking the preview point. An improved controller with the fuzzy control algorithm was designed to realize the compensation of the front wheel angle of agricultural machinery. Specifically, the lateral error and heading error of the agricultural machinery were used as the inputs of the fuzzy controller, and the compensation amount of the front wheel angle was used as the output. The sum of the feedforward and compensation amount of the front wheel angle were used as the control amount of the front wheel angle of the vehicle. This control amount considered the lateral error and heading error of the current position of the vehicle, together with the influence of the forward reference path on the steering of the agricultural machinery. The tracking accuracy was then greatly improved in the agricultural machinery. Taking the DF2204 continuously variable transmission tractor as the test platform, an automatic navigation system was developed to verify the practicability of the path tracking algorithm. 21 sets of variable curvature path tracking experiments were carried out in Beijing Miyun Experimental Field. The results indicate that the average absolute errors were 2.7, 2.7, 3.3, and 4.0 cm, and the average root errors were 3.4, 3.7, 4.6, and 5.0 cm, respectively, when the tractor moved at 1.0, 1.5, 2.0, and 3.0 m/s. The tracking control approach of variable curvature path in the autonomous navigation can be expected to effectively increase the tracking accuracy of agricultural machinery curve paths and the utilization rate of farmland.