Transit-MP: Transit-Prioritized Max-Pressure Control in Sparse Connected Vehicle Environments

Chaopeng Tan, Hao Liu, Dingshan Sun, Marco Rinaldi, Hans van LintarXiv 2025 · DOI · arXiv

This paper, recently published by researchers from TU Delft, TU Dresden, and the University of Maine, introduces Transit-MP and its modified version mTransit-MP. These algorithms evolve the classic "Max-Pressure" traffic signal control to prioritize public transit while maintaining network stability, even when only a small fraction of vehicles on the road are "connected."

TL;DR

The authors developed a decentralized traffic signal control system that prioritizes high-occupancy transit (buses and trams) using sparse data from connected vehicles. By accounting for transit dwell times and using historical data to fill real-time "blind spots," they successfully prevented network-clogging spillovers and reduced passenger delays by up to 18% compared to existing state-of-the-art methods.

What Problem Does This Paper Address?

Traffic signal control is moving toward a Max-Pressure (MP) paradigm. MP is popular because it is decentralized (each intersection makes its own decisions) and mathematically guaranteed to keep queues stable. However, standard MP has two major "blind spots":

  1. The Transit Gap: Most MP models treat a bus with 50 people the same as a car with one person. While some "occupancy-aware" models exist, they often ignore dwell time—the period a bus spends at a stop. Prioritizing a bus that is still boarding passengers wastes green time for everyone else.
  2. The Sparse Data Problem: MP models increasingly rely on Connected Vehicle (CV) data (GPS, speed, etc.). If only 10% of cars are connected, an intersection might "see" an empty road when it’s actually packed with non-connected vehicles. This leads to Queue Starvation: the signal never turns green because it thinks no one is there, causing a massive traffic jam (spillover) that the system is blind to.

Methods

The authors propose two iterations of their controller:

  • Transit-MP: This model uses occupancy-weighted link travel time. It calculates "pressure" based on how many passengers are delayed. Crucially, it introduces a binary parameter ($\beta_v$) that excludes transit vehicles from the pressure calculation while they are at a station. They only "count" toward requesting a green light once they have finished dwelling and are approaching the intersection.
  • mTransit-MP (Modified): To solve the "Queue Starvation" issue in sparse environments, this version incorporates historical traffic data. If no real-time CVs are detected, the system uses an Incremental Queue Accumulation (IQA) model based on historical arrival rates to estimate the current queue. This ensures the signal eventually turns green even if no connected vehicles are in line.

Theoretical Contribution: The authors provide a rigorous mathematical proof using Lyapunov stability theory to show that their system remains stable even when using different traffic measures for upstream and downstream links—a significant departure from standard MP proofs.

Datasets

The researchers evaluated their models using SUMO (Simulation of Urban MObility) on a high-fidelity digital twin of a real-world corridor in Amsterdam.

  • The Network: 3 signalized intersections, 15 transit lines (7 tram, 8 bus), and 31 stations.
  • Traffic Mix: A combination of dedicated transit lanes, mixed-use lanes, and general private vehicle lanes.
  • Scenarios: They tested varying CV penetration rates, from "sparse" (10%) to "full" (100%).

Key Insights and Ideas

  • Don't Prioritize the Dwell: The "aha moment" here is that transit priority should be conditional. A bus at a stop is a stationary object, not a traffic flow. By ignoring dwelling buses, the system "saves" green time for other movements without actually delaying the bus.
  • Historical Data as a Fail-Safe: In sparse CV environments, real-time data is a luxury, not a guarantee. Using historical averages to "nudge" the controller prevents the mathematical "zero pressure" state that causes queue starvation.
  • Stability with Asymmetry: Traditionally, MP requires symmetric measurements (e.g., queue length vs. queue length). This paper proves you can use occupancy-weighted travel time on one side and raw travel time on the other without the system "blowing up" (becoming unstable).

Key Results

  • Efficiency Gains: In a 100% CV environment, Transit-MP reduced passenger delay by 17.9% and private vehicle delay by 21.8% compared to the previous state-of-the-art (eOCC-MP).
  • Spillover Mitigation: Transit-MP almost eliminated link spillovers, reducing the maximum spillover count by 94.2% compared to earlier models.
  • Robustness in Scarcity: At only 10% CV penetration, the modified mTransit-MP was the only controller that prevented massive gridlock. It maintained stable vehicle counts while other models saw queues climb indefinitely due to "blindness" to non-connected cars.
  • Error Tolerance: The system proved resilient to estimation errors (up to 50% error in arrival rate/queue estimation), making it viable for real-world deployment where historical data might be imperfect.

Limitations and Future Work

  • Privacy: The model assumes the system knows the occupancy (number of passengers) of vehicles. While feasible for transit via APC (Automatic Passenger Counter) data, doing this for private cars raises privacy and technical hurdles.
  • Multi-Modal Complexity: The current model focuses on cars, buses, and trams. Future work needs to integrate pedestrians and cyclists, who have very different "pressure" profiles.
  • Link Segmentation: The authors noted that on very long roads, the "pressure" calculation can get diluted. They suggest that segmenting long links into smaller virtual blocks could further improve performance.

Reference: Tan, C., Liu, H., Sun, D., Rinaldi, M., & van Lint, H. (2025). Transit-MP: Transit-Prioritized Max-Pressure Control in Sparse Connected Vehicle Environments. arXiv:2511.00309v1 [math.OC].

Summary updated

Abstract

Max-pressure (MP) control stands out among real-time network traffic signal control methods due to its simplicity, decentralized nature, and theoretical stability. However, existing MP control methods have limited consideration of public transportation and do not address the network stability problem of transit-prioritized MP in partially connected vehicle (CV) environments. In this study, we propose Transit-MP, which realizes transit-prioritized MP control in partially CV environments by considering real-time vehicle occupancy and the impact of transit dwell at stations. Theoretically, we demonstrate that Transit-MP, while using different traffic state measures for upstream and downstream links for pressure calculation, still achieves road network stability even in partially CV environments. Note that for MP controllers in sparse CV environments, some movements may have missing CV observations, leading to link spillovers, which create the queue starvation phenomenon: a movement no longer receives the green phase despite the queue spillover. Therefore, we further propose a modified Transit-MP (mTransit-MP) that incorporates historical traffic data to address this issue. We rigorously prove that the proposed mTransit-MP can effectively avoid the queue starvation phenomenon. Experimental results on a real-world corridor in Amsterdam with 15 transit lines and 31 stations show that our method significantly reduces the real-time vehicle and spillover count, and improves delays for both private vehicles and transit vehicles compared to a state-of-the-art MP controller for transit signal priority. In sparse CV environments, our mTransit-MP is effective in mitigating link spillovers while enhancing the overall performance of multi-modal traffic.
BibTeX
@misc{tan2025transit,
  author = {Chaopeng Tan and Hao Liu and Dingshan Sun and Marco Rinaldi and Hans van Lint},
  title = {{Transit-MP: Transit-Prioritized Max-Pressure Control in Sparse Connected Vehicle Environments}},
  year = {2025},
  eprint = {2511.00309},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2511.00309},
  howpublished = {arXiv},
}
Figure