Abstract: Aiming at the issues of long search time, high path redundancy, and non-smooth paths in the traditional rapidly-exploring random tree (RRT) algorithm for robotic arm obstacle avoidance path planning, a Gaussian mixture model RRT (GMM-RRT) algorithm based on an improved Gaussian mixture model sampling strategy is proposed. Firstly, the algorithm constructs a guided sampling model by introducing multiple Gaussian distributions in space and combines random sampling to balance exploration and target orientation. Secondly, an adaptive dynamic step size is incorporated to enhance expansion efficiency and path quality in complex environments. Finally, a greedy pruning strategy removes redundant points from the initial path, and the pruned path is smoothed using cubic B-spline curves. To evaluate the improved algorithm, three experimental environments were set up in Matlab, and comparative simulations were conducted with three algorithms including the improved RRT. The results show that in a complex 3D environment, the improved RRT algorithm reduces path length and the total number of nodes by 30.27% and 84.32%, respectively, while significantly shortening planning time. The feasibility of the improved algorithm was verified through six-axis robotic arm 3D obstacle avoidance path planning experiments, demonstrating its ability to provide safer and smoother paths and improve operational efficiency.