A Hierarchical Path Planning Algorithm for Autonomous Ground Vehicles Based on Q-learning and MPC

In practical applications, such as item transportation, motion planning and control of autonomous ground vehicles become challenging due to the uncertain operating environment. In this paper, put forward a control scheme that finds a better path by combining Q-learning autonomous ground vehicles path planning with artificial potential field and path optimization based on model predictive control. The grid method is commonly used to describe the environment for path planning based on reinforcement learning algorithm, but there are some problems such as paths that are too close to obstacles and non-shortest paths, which are not consistent with the actual application scenarios. To solve this problem, a Q-learning autonomous ground vehicles path planning algorithm combined with artificial potential field knowledge is proposed. The repulsion field value of obstacles is introduced to optimize the reward value when selecting the state, and the oblique motion of autonomous ground vehicles is increased. The simulation results indicate that, compared to the original algorithm, the Q-learning autonomous ground vehicles path planning algorithm combined with artificial potential field (APF) can find a better path that better aligns with reality when the time consumption is increased. However, since the planned path in the upper layer does not conform to the motion principle of autonomous ground vehicles, the model predictive control (MPC) is used to optimize planned path. The results demonstrate that proposed method delivers satisfactory motion planning and control.