Abstract: Aiming at the problems of low planning efficiency, redundant iteration times and high node redundancy existing in the random tree expansion-based RRT algorithm for manipulator path planning, an improved path planning algorithm based on a three-tree cooperative growth mechanism is proposed. The improved algorithm adopts a three-tree cooperative architecture consisting of a start-connected tree, a central expansion tree and an end-connected tree, synchronously constructs the start point-connected tree and the end point-connected tree, and realizes bidirectional cooperative convergence towards the central tree. An adaptive limited sampling space is designed and fused with the target-biased expansion technology to significantly improve the effectiveness of tree node generation and the directionality of tree growth. After path generation, reverse search pruning is used to complete the rough path optimization, and forward interpolation pruning is combined to achieve precise path optimization. The cooperative effect of the two shortens the path length, and finally, the cubic B-spline curve is used to complete the path smoothing process. Simulation experiments on 3D random maps based on Matlab and MoveIt2 of the ROS2 platform show that, compared with traditional algorithms, the improved algorithm shortens the planning time by 40% to 70% and reduces the number of iterations by 50% to 70% on the premise of a slight reduction in path length, which effectively improves the efficiency and quality of manipulator obstacle avoidance path planning.