Abstract
Autonomous vehicles in structured spaces like parking lots face challenges due to limited space, obstacles, and motion constraints. Traditional methods like A* are slow, and basic RRT or Dubins paths lack flexibility. This study improves 2D path planning by enhancing RRT* with spline fitting, node pruning, and two phase sampling and compares its performance with Dubins and Reeds-Shepp models for smooth, collision-free, and realistic navigation. The data used in this study was generated through simulation in three structured environments: a compact parking lot, a parallel parking setup, and a mall-style parking area. Algorithms like RRT, RRT* Dubins, and Reeds-Shepp were tested using Python tools (Pygame, NumPy, Matplotlib) and MATLAB for validation. Each setup involved 10 test runs with fixed start and goal points, and metrics such as path length, time, and success rate were recorded to evaluate algorithm
performance.
The study found that Reeds-Shepp paths combined with RRT* performed best in structured, tight environments due to their support for reverse motion. RRT* generated smoother, more optimal paths than standard RRT, especially when enhanced with spline fitting and node pruning. Dubins paths were faster in open areas but less effective in compact spaces.
Overall, Reeds-Shepp with RRT* was the most practical for real-world autonomous parking.