Abstract: Peanut harvesting has posed a great promise for the sustainable development in modern agriculture. This study aims to improve the path-tracking accuracy of an unmanned crawler-mounted peanut harvester on sandy land. 4HBL-2 self-propelled peanut combine harvester was taken as the research object. A systematic investigation was also carried out on the unmanned path-tracking control of the crawler-mounted harvester. The optimal relationship was established between the kinematic model of the crawler-mounted harvester and the virtual steering angle function. The course deviation was used as the observation value, whereas, the angular velocity calculated by the Ackerman model was used as the measurement value. Kalman Fusion Algorithm was also designed to obtain the virtual steering angle using the Ackerman model. The PID path tracking was improved significantly, according to the virtual steering angle. A double PID path tracking control was proposed using preview tracking. A pulse width controller was then selected to realize the accurate path-tracking control of the crawler-mounted peanut harvester. The simulation test results showed that the path tracking control method based on preview tracking double PID can perform path tracking control, and had the characteristics of smooth control and small steady-state error. There was no change in the signal period and waveform distortion. A series of field experiments show that the average absolute error and the maximum deviation of the linear tracking were 2.23, and 4.14 cm, respectively, when the peanut harvester was operated at a speed of 0.6 m/s in the sand. The performance of the improved system was enhanced by 56.12%, and 66.07%, respectively, compared with the PID path tracking controller. The path tracking experiments showed that the response duration values of the control system were 11.00, 12.92, and 13.78 s, respectively, while the corresponding distances were 6.60, 7.75, and 8.26 m, respectively, when the initial deviation was 0.5, 1.0, and 1.5 m, respectively. Specifically, the maximum overshoot was 5.68, 5.84, and 6.09 cm, respectively. The tracking path distance was reduced by 1.92%, 4.43%, and 8.71%, respectively, whereas, the overshoot was reduced by 8.45%, 17.56%, and 5.17%, respectively, compared with the wheeled harvesters. The maximum deviation and average absolute error of the peanut harvester were 5.87 and 2.72 cm, respectively. The ratios of deviation within ±5 cm and ±10 cm were 97.11%, and 100%, respectively. The general technical conditions were also compared for the post-installation of the satellite navigation automatic driving system of agricultural machinery in the industry standard. The connection performance of the crawler peanut harvester can be expected to meet the high requirements of the connection performance. The path tracking control accuracy of the crawler-mounted peanut harvester can also fully meet the harsh requirements of unmanned operation in sandy land.