In order to improve the navigation efficiency of the kiwifruit harvesting robot, the rapidly expanding random tree (Rapidly-exploring random trees, RRT) algorithm was researched, and an improved method (Straight-RRT) for real-time guided random tree expansion based on sampling state was proposed. Firstly, aiming at the problem of blind search in the traditional RRT algorithm, an evaluation index and a threshold were introduced to divide the sampling state, and the selection method of sampling nodes was determined according to the sampling state, so as to guide the expansion of the random tree in real time. Secondly, in order to enhance the adaptability of the algorithm to different environments, a dynamic threshold was introduced, which was adaptively adjusted according to the complexity of the environment, and at the same time, the nearest node selection mechanism was optimized to make the random tree avoid irregular obstacles faster. Finally, the path was optimized, the redundant points of the path were removed, and the Bezier curve was used to smooth the path to reduce the complexity of the path. Combined with the environment of the kiwifruit orchard, the effective harvesting area of the kiwifruit harvesting robot was analyzed, and a navigation method to achieve full coverage of the kiwifruit orchard was proposed. Based on the kiwifruit orchard environment, the path planning experiment was carried out, and the algorithm was compared with other improved algorithms. The path planning experiment results showed that the improved algorithm had better adaptability and planning efficiency in the kiwifruit orchard environment, which can provide a basis for improving the navigation efficiency of the kiwifruit harvesting robot.