Implementation of Digital Virtual Environment Model ConsideringObstacles and Traffic Lights, and Research on Multi-LaneAutonomous Driving Based on Deep Reinforcement Learnin

초록

In this paper, a self-driving system in a digital virtual road environment utilizing deep reinforcement learning is proposed. Using ML-Unity, a digital virtual environment is created to simulate a multi-lane roadwith various obstacles and traffic lights. Multiple sensors are deployed on the vehicle to observe the current road and driving environment, facilitating the development of the autonomous driving system. Information about obstacles, traffic lights, and surrounding vehicles is acquired through the digital virtual environment. This information is then mapped to the state space of the deep reinforcement learning model to dynamicallydetermine actions, such as driving direction and speed, to maximize performance in terms of driving distanceand time. The paper introduces a system design that combines priority experience replay-based DeepQ-Network (DQN) with exploration strategies and a novel reward function to achieve fast learning and stabledriving. Through experiments in the digital virtual space, the proposed system is validated to successfullyperform lane-keeping, obstacle avoidance, and compliant driving with traffic signals compared to vanilla DQN. © 2024, Korean Institute of Communications and Information Sciences. All rights reserved.

키워드