Abstract: Intelligent and unmanned agricultural machinery can be very necessary to develop in modern agriculture, due to the shortage of labor force against the urbanization in recent years. It is also in high demand to enhance the utilization of land and the efficiency of the machinery. Among them, autonomous driving can be expected to serve as the key technology for unmanned agricultural machinery. In this study, a set of additional auto-driving systems was established to fully meet the requirement of intelligent improvement on traditional harvesters in the full-field blocks. Firstly, a hand-compatible electric control device was designed for the simple switch between manned and unmanned operations in current agricultural production, according to the control and power system of the traditional harvesters. Secondly, a series of experiments were conducted to test the measurement and control system, as well as the steering characteristics of the single-sided brake steering harvester. The operation performance of the controller was verified using the response of the actuator, together with the relationship between steering valve opening, HST speed, and steering curvature. PD-fuzzy-BangBang joint control was then proposed to improve the accuracy of the real embedded control system, according to the pedrail steering, measurement, and control system. Finally, a prototype was developed for the automatic driving and operation on the cement surface in the rice field. The full-block automatic driving performance was achieved in the joint control system of the prototype. The cement surface online experiment showed that the over-adjustment oscillation of the system was effectively reduced for the steady-state accuracy on the line. Specifically, the online distance of the combined algorithm was shortened by 57.3%, and the steady state standard deviation was reduced by 81.3%, compared with the single PD. The cement ground experiment showed that the prototype better performed the full-block automatic driving under ideal conditions on straight paths, with a maximum deviation of 6.00 cm and a standard deviation of 2.42 cm. The accuracy of the prototype fully met the requirements during actual operation in the rice field, particularly with the over 80% cutting width and the speed of 0.7 m/s. Therefore, full-field unmanned harvesting was realized with the addition of the self-adaptive operation function. The finding can provide technical support and equipment solutions for the construction of unmanned farms.