2D Maze Path Planning Based on Q-Learning and Bezier Curve

Q-Learning is widely used in path planning situations, while the trajectories it generates usually have abrupt bends, bending points, and sharp endpoints. A modified Q- Learning algorithm and Bezier curve are introduced in this study to improve trajectory smoothing, bringing the resulting trajectory closer to the routes of real-world vehicles. First, different mazes with the same start point and end point are created. Second, Q-Learning and modified Q-Learning are applied to path planning. Lastly, the path smoothed by the Bezier curve is imported into the simulation environment to verify its validity. The results demonstrate the effectiveness of the above methods in enhancing path smoothness and shortening path length.