In view of such flaws as poor expandability, multiple redundant nodes, and poor path quality of RRT（rapidly-exploring random tree） in the path planning process of manipulators, an algorithm based on AGD-RRT (adaptive goal-directed RRT) has been proposed to address these above-mentioned problems. Firstly, the algorithm constructs a dynamic goal-directed probability function that adjusts the sampling probability of target points in real-time to achieve adaptive target orientation, thus reducing the generation of useless nodes and improving convergence speed. Secondly, a greedy convergence strategy is adopted for the prevention of a blind expansion of the random tree around the target. With the search completed, the node removal method is used to remove redundant nodes in the path, with the trajectory removed by B-spline curves for an improvement of the quality of the path. Then, comparative simulation experiments are to be conducted in 2D and 3D environments to verify the feasibility and superiority of the algorithm. Finally, a prototype experiment is conducted to verify the feasibility of the proposed algorithm for path planning in the joint space of the manipulator.