Abstract: The generation of power shift tractor promotes the automatic driving of agricultural machinery to the direction of unmanned, and the automatic obstacle avoidance of agricultural machinery has become a key issue to be solved. Most of the existing obstacle avoidance path planning adopts the shortest tangent method, which has the advantages of fast and simple path generation and short track, but the curvature of the path curve is discontinuous and there are corner inflection points, which make the tractor front wheel angle suddenly change and affect the obstacle avoidance accuracy. In this paper, for the curvature discontinuity of the shortest tangent path curve, poor tracking control accuracy and low precision of agricultural machinery model, the third-order bezier curve optimization method was used to form a continuous smooth agricultural machine obstacle avoidance path. The optimized curve curvature was continuous and the obstacle avoidance path was smooth with no turning points, which was conducive to the control of agricultural machinery driving. The linear control model of agricultural machinery movement was established by chain control theory, and the PI controller was used for corner compensation to improve the accuracy of agricultural machinery obstacle avoidance. The simulation of the path tracking control method was completed. The simulation results showed that the heading error angle of the agricultural machinery was mostly in the range of -0.06-0.06 rad, the lateral position error was less than 13 cm, and the front wheel steering angle changed gently with no significant mutation, indicating that the control method of the obstacle avoidance path control method had a high precision, and the agricultural machinery can drive according to the preset trajectory. The ploughing operation test was carried out, and the Dongfanghong LF954-C power shift tractor was used, which can automatically start, stop, shift, and implement the lifting operation of the farm. The tractor was equipped with the Beidou agricultural autopilot with the straight running accuracy of ±2.5 cm. Ultrasonic radar and image recognition technology were used to perceive obstacles in the working environment. The farm implement was a 250 cm wide plow. A haystack obstacle was set on the machine working path with a contour radius smaller than the minimum turning radius of the tractor. The results of the plowing operation test showed that the machine can automatically complete the straight-line driving, farm implement lifting, automatic obstacle avoidance, farm implement falling, and continuous straight-line operation. The obstacle avoidance accuracy of the bezier curve was 5.21 cm, and the tracking control effect of the curve path was well. The accuracy of the agricultural machine continuing to travel in a straight line after obstacle avoidance was 1.98 cm, which indicated that the path tracking control method can ensure the tractor straight-line automatic driving after obstacle avoidance. The above results showed that the obstacle avoidance path and control method showed good robustness and adaptability in uneven ploughing land and can meet the obstacle avoidance requirements of tractor operation.