Trained Model Reuse of Autonomous-Driving in Pygame with Deep Reinforcement Learning

Abstract

Autonomous-Driving technology has begun to bring great convenience to daily trip, transportation, and surveying harsh environment. Considering that deep reinforcement learning has requirements for the convergence performance of the training results, and the actual training results sometimes cannot converge steadily or fail to reach the training goals, in this paper, the trained model reuse method was proposed, which can use the trained model generates Q(St, At) and can be used as a part of Deep Reinforcement Learning model, and this model was built based on the value function that could predict the Q value corresponding to the various actions performed in the environment state. In the Pygame platform, a simplified traffic simulation environment was set, it is observed that the Autonomous-Driving vehicle could run smoothly without collision in a fixed-length test simulation environment, and this trained model reuse method could help autonomous vehicle accelerate the learning process, obtain better simulation results during most of the training process, save simulation time and computing resources.