Transportation

Deep reinforcement learning for decision making of autonomous vehicle in non-lane-based traffic environments

Y. Fei, L. Xing, et al.

Using toll-plaza diverging areas as a case study, this study builds a microscopic simulation of realistic human-driven trajectories and proposes a deep reinforcement learning–based lateral motion strategy for autonomous vehicles with tailored state and reward functions. Simulations show reduced single-vehicle diverging time and that moderate AV penetration improves efficiency and safety while excessive penetration harms operations. Research conducted by Yi Fei, Lu Xing, Lan Yao, Zhizhi Yang, and Yujie Zhang.