Human-Like Behavior Decision Making in Autonomous Driving Based on Deep Reinforcement Learning

Autonomous driving is a key focus for future transportation systems, where decision-making in dynamic environments is critical. Current models have difficulty ensuring the safety of decision-making. To address this problem, a humanlike autonomous driving decision-making model is proposed, which incorporates driver style into the deep Q network (DQN) algorithm. First, by simulating different driving styles, the model can optimize decision-making, improve safety, and reduce collision rate. Then, the design of the reward function reflects the characteristics of the driver, and the <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{Q}$</tex> function weights of DQN are adjusted to consider goal achievement, speed maintenance, safe distance, and time efficiency in highway scenarios. Simulation experiments further verify its potential in optimizing highway driving decisions. And compared with the proximal policy optimization (PPO) algorithm, the proposed model reduces the collision rate by 6.33 % in aggressive driving scenarios.