Application Research of end to end behavior decision based on deep reinforcement learning

At present, the control method of driverless vehicle mainly adopts artificial rule making and behavior decision-making, which is difficult to adapt to the new scene. To solve this problem, this paper proposes an end-to-end behavior decision-making method, which uses deep reinforcement learning to interact with the environment and learn the decision model. The model has good generalization. The main reinforcement learning algorithms used in this paper are PPO, SAC, A2C, DDPG and TRPO algorithm, etc. This paper is based on the unity 3D virtual engine to build the scene of vehicle search target. The experimental results show that DDPG algorithm has better effect, the better generalization performance, the shortest time to find the target and the fastest training speed.