강화학습 기반 CAV 거동 제어 알고리즘을 통한 통제차로 회피 전략 개발 및 평가

초록

Autonomous vehicles (AVs) are expected to positively impact traffic flow by eliminating human factors and ensuring more ideal and consistent driving behaviors. However, due to the limitations of vehicle-centric autonomy, AVs still fail to appropriately handle various unexpected situations, adversely affecting traffic flow and causing crashes. Therefore, for AVs to achieve their anticipated positive impact on traffic flow, it is necessary to evolve into Connected Autonomous Vehicles (CAVs), including the cooperation of infrastructure based on Vehicle to Everything (V2X) communication. The cooperative driving effect of CAVs will be evident in the case of lane closed situations caused by unexpected events such as traffic accidents or construction. If CAVs can acquire information about the lane closed situation through infrastructure communication in advance and control their maneuvers of CAVs appropriately, it is expected that the systematic movement of CAVs will alleviate issues such as bottlenecks caused by closed lanes and ensure a smoother traffic flow. This indicates the necessity for research on strategies to smoothly avoid and pass through closed lane situations in future LV4 autonomous driving situation where CAV maneuver control is possible. Reinforcement learning, which can accommodate large state spaces like traffic situations, has recently emerged as a promising framework for addressing the control challenges of CAVs, which must flexibly respond to real-time dynamic traffic conditions. Therefore, this study developed a reinforcement learning-based strategy for avoiding closed lanes and evaluated its impact on traffic flow, thereby demonstrating the utility of reinforcement learning as a methodology for developing closed lane avoidance strategies for CAVs in future connected environments.

키워드