<p>Path-planning algorithms play a crucial role in improving the operational efficiency of robots. However, achieving efficient path planning in high-dimensional spaces, particularly for manipulators, is still a challenge. To address the limitations of the conventional rapidly exploring random tree (RRT) algorithm, such as low search ability, poor adaptability to complex environments, high randomness, and tortuous paths, this paper proposes an improved RRT algorithm for obstacle avoidance path planning of manipulators. Firstly, a kinematics model of the manipulator is established, followed by forward kinematics, and the reachable workspace is evaluated by Monte Carlo method. Then, an adaptive tree expansion and sampling strategy is introduced, guided by collision feedback, to effectively balance global exploration and local guidance. An initial path is rapidly generated through a global adaptive step-size strategy, and a smooth, feasible trajectory suitable for the manipulator’s motion is obtained via path optimization. To evaluate the performance of the proposed algorithm, three different environments, including both 2D and 3D scenarios, were constructed in MATLAB. The improved RRT was compared with the standard RRT, Bi-Directional RRT (Bi-RRT), Goal-Directed RRT (G-RRT), improved G-RRT, and Restricted Sampling Area RRT (RSA-RRT). Simulation results show that the improved RRT outperforms these algorithms, with lower average running time, iterations, path cost, and other performance metrics. In addition, the feasibility and practicality of the proposed algorithm are further validated through path planning and obstacle avoidance experiments conducted both in MATLAB and on a physical manipulator.</p>