In China, the village roads are characterized by numerous intersections and significant differences in road widths, creating a complex maze-like terrain. This undoubtedly increases the difficulty of path planning for autonomous vehicles. This study proposes an improved bidirectional RRT* algorithm that utilizes the advantages of the rapid random search of the RRT* algorithm. It introduces virtual points to address the irregularity of road networks and creates enveloping circles at expanding nodes to enhance path reachability, thus obtaining the optimal global planning path. To enhance path tracking comfort, a fifth-order B-spline curve is utilized to smooth the global path, and local path planning is performed using Quadratic Programming (QP). The proposed combined global and local path planning method was evaluated through Co-simulation experiments basing on the Matlab/CarSim/PreScan platform. Simulation results demonstrate that the enhanced RRT* algorithm outperforms the traditional RRT* algorithm in the same scenario. Specifically, the improved algorithm reduces the running time by 29.56%, increases node utilization by approximately 15.33%, and decreases the planned path length by 2.8%. Additionally, the vehicle’s final lateral tracking error was controlled within 0–0.04 m, and the longitudinal tracking error was controlled within 0–0.1 m, fully demonstrating the vehicle’s excellent path-tracking performance. The study’s innovative ideas will offer methodological support for researching path planning for autonomous vehicles in specific scenarios.