In this paper, we present a novel autonomous peach-harvesting approach. The robot can autonomously move around the tree and detect the peaches accurately by a well designed fully connected neural network, adapting to different illumination. Next, the robot generates a score for each peach, taking into account the confidence of the classification and the size of the bounding box, and then pick one of the most suitable peach. In order to track the peach more robustly and quickly, we choose the KCF tracking algorithm. Finally, the robot picks the

peach through a 6-DOF manipulator. In addition, we propose a monocular high-precision ranging method based on structured light. The cost of our method is lower than the traditional

ranging method, which enables the robot to obtain the accurate spatial position of the peach.