Liu, Bowen (Tsinghua University), HU, Jianming (Tsinghua University), Jiang, Xingwei (Tsinghua University)

MixLight: A Model-Based Reinforcement Learning Method for Traffic Signal Control in Mixed Traffic with Low Connected Vehicle Penetration

Scheduled for presentation during the Video Session "On-Demand Video Presentations" (VP-VP), Saturday, November 22, 2025, 08:00−18:00, On-Demand Platform

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 April 2, 2026

Abstract

The rapid growth of Telematics and autonomous driving technologies has significantly transformed urban traffic systems, which will evolve into mixed flows comprising Regular Vehicles (RVs) and Connected Vehicles (CVs). Existing traffic signal control methods often require high CV penetration rates to optimize signal timing, limiting their applicability in low-CV scenarios. We propose MixLight, a model-based reinforcement learning framework designed to address the complexities of mixed traffic flow environments, where both CVs and RVs coexist. MixLight introduces three core innovations: (1) a compact vector representation that encodes spatiotemporal traffic states through a multi-head attention autoencoder, allowing low penetration scenarios to be trained with information from complete traffic scenarios; (2) a temporal transition model for predicting future traffic states, enabling proactive control with minimal real-time interactions; and (3) a dual-phase training strategy that jointly leverages complete and partial traffic observations, ensuring robustness under varying CV penetration rates. Experiments on synthetic and real-world datasets demonstrate that MixLight outperforms traditional methods and other DRL-based methods. This work provides a scalable and data-efficient solution for traffic signal control in mixed traffic flows.