Transit Signal Priority Control With Connected Vehicle Technology: Deep Reinforcement Learning Approach

Fan, Wei; Yang, Tianjia · 2024 · ROSA P / University of North Carolina at Charlotte. Center for Advanced Multimodal Mobility Solutions and Education

archive: archived pipeline: cataloged verified

Get this paper ↗ (full text — opens at the source; we link to it, we don't host it)

Summary

This study addresses the challenge of optimizing Transit Signal Priority (TSP) to improve public transportation efficiency without causing excessive delays for private vehicles. While TSP is a critical strategy for sustainable urban mobility, traditional implementations often negatively impact general traffic, limiting their adoption. The authors leverage Connected Vehicle (CV) technology, which provides real-time, fine-grained data such as passenger occupancy, to enable more sophisticated, adaptive control. The research aims to develop robust, data-driven TSP controllers using Deep Reinforcement Learning (DRL) that can handle dynamic traffic conditions at both isolated intersections and multi-intersection corridors. The methodology employs two distinct DRL frameworks tailored to different spatial scales. For isolated intersections, the study utilizes Single-Agent Reinforcement Learning, specifically Deep Q-Networks (DQN) and Proximal Policy Optimization (PPO), formulated as Markov Decision Processes. For corridor-level control, it applies Multi-Agent Reinforcement Learning (MARL), specifically Multi-Agent PPO (MAPPO), formulated as Decentralized Partially Observable Markov Decision Processes. These controllers were trained and evaluated in simulation environments using both hypothetical and real-world traffic configurations. The performance metrics focused on average person delay, allowing the system to prioritize buses with higher occupancy while balancing overall network efficiency. The results demonstrate that the proposed DRL controllers significantly outperform traditional pretimed and actuated signal controllers. In isolated intersection scenarios, the DQN-based controller reduced average person delay by 18.77% during peak hours and 23.37% during off-peak hours compared to pretimed signals, while also decreasing delays for both buses and cars. Sensitivity analyses confirmed the controller’s robustness against variations in CV market penetration, bus occupancy, and arrival headways. In corridor scenarios, the MAPPO-M controller, which utilizes a multi-discrete action space, achieved superior bus service performance while maintaining acceptable service levels for regular traffic. The study found that MAPPO-M performs optimally when CV market penetration exceeds 60% and effectively adapts to dynamic changes in passenger loads and bus schedules. The significance of this work lies in its demonstration that DRL, combined with CV data, can resolve the trade-off between transit priority and general traffic efficiency. By shifting the optimization objective to person delay rather than vehicle delay, the proposed frameworks encourage public transit usage by reducing passenger travel times. The findings suggest that these adaptive controllers are viable for practical implementation, offering a scalable solution for modernizing traffic signal infrastructure. This approach supports the development of more equitable and efficient urban transportation systems by leveraging emerging connectivity technologies to enhance multimodal mobility.

Key finding

Deep reinforcement learning-based transit signal priority controllers significantly reduce average person delay and improve transit service compared to traditional pretimed signals, with corridor-level performance optimizing when connected vehicle penetration exceeds 60%.

Methodology

simulator

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed. Discovered via bulk_ingest_rosap on 2026-05-23 (6 acquisition events logged).

Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• teleoperation remote driving

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Theoretical Contribution: computational model