Abstract: Due to the uncertainty and complexity of the harvesting environment for fruits and vegetables, the manipulator needs to consider real-time obstacle avoidance in completing harvesting tasks in complex environments. In order to achieve safe harvesting of manipulator in uncertain environments, an improved dynamic obstacle avoidance algorithm based on the rapidly-exploring random trees（RRT） algorithm is proposed to enhance the adaptability of manipulator in uncertain harvesting environments. In order to address the issues of long iteration time, long path length, and poor adaptability in dynamic environments of the basic RRT algorithm, this study first introduces a target-oriented strategy to increase the iteration efficiency of the algorithm by randomly sampling points with a certain probability towards the endpoint. Secondly, a dynamic detection mechanism is introduced to dynamically detect the initial path that has been planned, making the algorithm adaptable to changes in the environment. Simulation analysis shows that the improved RRT algorithm reduces path length by 16% and shortens iteration time by 86.5% compared to the basic RRT algorithm. Furthermore, the dynamic detection mechanism allows the algorithm to adapt to dynamic environments.