To address the issues that parameter adjustment in the traditional A* algorithm for path planning relies on experience and it is difficult to balance planning efficiency and quality, an improved A* algorithm incorporating a radial basis function (RBF) neural network was proposed. This algorithm utilized the RBF neural network to adaptively adjust the weight coefficient α of the A* algorithm based on environmental features, enabling autonomous parameter optimization. Meanwhile, the traditional eight-neighborhood search mechanism was optimized into a four-neighborhood expansion and combined with a path smoothing strategy, enhancing search efficiency while improving path executability. Simulation results demonstrate that the improved algorithm reduces the path length by 26.0% compared to the traditional four-neighborhood method and decreases the planning time by 91.1% compared to the eight-neighborhood method. Furthermore, real-world flight experiments conducted on the robot operating system (ROS) platform validate the method’s stable navigation and effective obstacle avoidance capabilities in complex obstacle environments.