Cui, Yunkuan (Tsinghua University), Yao, Danya (Tsinghua University), Ni, Xinrui (Tsinghua University), ZHANG, Yi (Tsinghua University), Zhou, Wei (Hebei Expressway Group Co., Ltd. Jingxiong Branch), Qi, Yuliang (Hebei Expressway Group Co., Ltd. Jingxiong Branch), Zheng, Ziyue (Hebei Expressway Group Co., Ltd. Jingxiong Branch), Zhang, Hongjun (Hebei Expressway Group Co., Ltd.)

Safety Improvement of Car-Following for Connected Vehicle Flow Based on Individual Speed Guidance Upstream of a Lower Speed Limit Expressway Section

Scheduled for presentation during the Regular Session "S40c-Cooperative and Connected Autonomous Systems" (FR-LA-T40), Friday, November 21, 2025, 16:20−16:40, Cooleangata 4

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 18, 2025

Abstract

The deceleration process of traffic upstream of a fixed lower speed limit bottleneck is one of the important situations causing the major accidents, rear end collisions, on expressways. Vehicles’ connectivity and cooperation can mitigate these risks by providing individual speed guidance based on their unique motion states. An environment model with a minimum duration value is built considering the acceptability of individual suggested speeds for approaching connected vehicles. A timely updated prediction control framework is developed to make the optimal control target of speed harmonization clearer and more intuitive. As this optimization problem is non-linear and high-dimensional, deep reinforcement learning is used to help get a feasible solution. The state space includes the suggestion speed at the last step and its duration, and the reward function includes the gap to the prediction control target in the framework. Training and validation simulation experiments show that the strategy can make a specific number of vehicles upstream driving through the deceleration section smoother and safer without much loss in efficiency by flexibly adjust their suggested speeds. A series of random validation scenarios show that the average and standard deviation of the related safety surrogate measurements over those scenarios are both improved a lot, particularly for those with high risks originally. At last, the analysis of the effects of different reward designs demonstrates it is necessary for our prediction framework to be included in the optimization process.