Aiming at the problems of low accuracy and slow convergence of rapidly-exploring random trees star fixed nodes(RRT*FN) algorithms, this paper proposes an improved RRT*FN path planning algorithm, which is used to solve the global path planning problem of mobile robots in the two-dimensional static environment. Firstly, the improved algorithm uses the same uniform sampling method as the RRT*FN algorithm for path planning. When the improved algorithm gets an initial path, the heuristic sampling method will be used in subsequent path searching. In each next iteration, the improved algorithm randomly selects a method from the ellipse subset sampling method and the path point neighboring area sampling method as the current sampling method. Then, when the total number of nodes in the tree reaches the preset value, the leaf nodes in the tree are deleted using a weighted method. By giving higher weight to the leaf nodes in the sampling area, the leaf nodes outside the sampling area are deleted with a higher probability. Therefore, high-performance nodes in the tree can be retained, and algorithm performance is improved. Finally, the simulation experiments verify the effectiveness of the improved algorithm.