Road Intersection Path Planning Based on Q-learning for Unmanned Ground Vehicle

This paper concentrates on the road intersection path planning problem of unmanned ground vehicle (UGV). First, the interaction between UGV and environment is established as a Markov decision process (MDP) model. Considering the feasibility of path, the kinematic model is also utilized to update the states of UGV, such as position, velocity and attitude. Then, the optimal driving strategy and path are generated by Q-learning algorithm. Reward function is designed to reflect the gain and loss of the chosen action. Finally, simulations demonstrate the feasibility of Q-learning in path planning of UGV.