Abstract: To address the problems that the picking robot arm employing the traditional fast random extended tree (RRT) algorithm requires a long time to search the path in the orchard working environment and the final path is not smooth with many inflection points, this paper proposes an improved RRT obstacle avoidance algorithm adopting Gaussian sampling strategy, which reduces the randomness of sampling, avoids generating more unnecessary random trees, and increases the orientation of programming. Then, A* cost function is added to remove redundant points of the path. Finally, greedy algorithm is used to simplify the path and reduce its inflection points, so that the robot arm can move along the optimal path quickly and locate the target point accurately. Our simulation results show the improved algorithm effectively reduces the time of path planning and cuts the path lengths, demonstrating its fairly good effectiveness and practicality.