Deep Q-Network (DQN) Approach for Automatic Vehicles Applied in the Intelligent Transportation System (ITS)

This paper presents the design of an intelligent controller applying reinforcement learning using a deep Q-network (DQN) algorithm for autonomous vehicles. The deep Q-network (DQN) algorithm is an online, model-free reinforcement learning approach. A DQN agent is a value-based reinforcement learning agent that teaches a critic to predict future rewards or returns. Deep Q-network is to replace the action-state Q table with a neural network. This solution applies to building a self-propelled agent capable of correcting static and moving obstacles according to the physical environment. As a result, the autonomous vehicle can move and avoid collisions with obstacles. The correctness of the theory is demonstrated through MATLAB simulation.