Automated Planning for Generating and Simulating Traffic Signal Strategies

Saumya Bhatnagar, Rongge Guo, Keith McCabe, Thomas L. McCluskey, Francesco Percassi, Mauro VallatiIJCAI 2023 · DOI

TL;DR

This paper introduces a novel framework that leverages automated planning to generate and simulate traffic signal strategies for urban traffic control. By integrating with existing traffic management systems like SCOOT, it allows for the testing and comparison of new strategies in a simulated environment before real-world deployment, addressing the challenges of dynamic traffic patterns in the post-pandemic era.

What Problem Does This Paper Address?

The COVID-19 pandemic has significantly altered travel patterns, making traditional, static approaches to urban traffic control insufficient. Existing systems struggle to adapt in real-time to unexpected fluctuations in traffic flow, necessitating more dynamic and responsive solutions. While AI techniques, particularly automated planning, show promise, they require robust models of traffic dynamics. Furthermore, traffic simulators, crucial for testing and comparing strategies, are often costly, limited in scope, and require extensive calibration, making them impractical for rapidly changing urban environments. This paper aims to bridge this gap by using automated planning to generate and simulate traffic signal strategies, utilizing existing traffic data and models.

Methods

The core of the proposed solution is a framework that integrates automated planning with traffic simulation. The system utilizes a module called SCOOT2Plan to process data from existing SCOOT (Signal Control by Optimized Output) systems. SCOOT2Plan extracts two key pieces of information: the current state of the network (initial state) and the historical traffic signal strategy implemented by SCOOT. This information, along with a dedicated AI Planning Domain Model written in PDDL+ (Planning Domain Definition Language), is fed into an AI planning system.

The PDDL+ domain model is crucial as it formally represents the dynamics of urban traffic control, including junctions, road links, traffic signal stages, intergreen times, and vehicle flows (turn rates). This declarative representation allows the planning system to generate optimal traffic signal strategies. The framework also includes a visualiser that allows for the comparison of simulated traffic signal strategies (both historical SCOOT plans and newly generated ones) against real-world data. The planning engine used is ENHSP, which can handle both strategy generation and simulation.

Datasets

The paper utilizes real-world data collected from sensors deployed in a major corridor of the Kirklees region in the United Kingdom. This data is processed by the SCOOT system and its associated database (ASTRID). For demonstration purposes, historical data from January 2022 was used, specifically focusing on a weekday (Wednesday, January 26th) and a weekend day (Saturday, January 30th), selecting periods with minimal disruptions.

Key Insights and Ideas

  • Bridging the Gap Between Planning and Simulation: The framework effectively combines the strengths of automated planning (generating optimal strategies) with the necessity of simulation (testing and validating these strategies) in a unified system.
  • Leveraging Existing Infrastructure: By integrating with SCOOT, the system can utilize existing traffic sensor data and infrastructure, reducing the cost and effort of deploying new systems.
  • PDDL+ for Expressive Traffic Modeling: The use of PDDL+ allows for a rich and declarative representation of complex traffic dynamics, including continuous variables and temporal aspects, which is essential for accurate simulation and planning.
  • Testing "What-If" Scenarios: The framework enables the simulation of novel traffic signal strategies that were not previously in operation, allowing traffic authorities to explore potential improvements and their impact before real-world implementation.
  • Post-Pandemic Relevance: The ability to adapt to and simulate dynamic, unpredictable traffic patterns makes this approach particularly relevant in the current urban mobility landscape.

Key Results

The paper presents a framework and demonstrates its capabilities. While specific quantitative results like percentage improvements in traffic flow or reduction in delays are not detailed in this demonstration-focused paper, the key qualitative findings are:

  • The framework successfully generates traffic signal strategies and simulates their impact on an urban region.
  • The generated strategies can be compared with existing SCOOT-implemented plans using a visualiser for validation and verification.
  • The system is capable of simulating new traffic signal plans to assess their effectiveness.
  • The approach is scalable to larger and more complex urban regions.

Limitations and Future Work

  • Focus on a Single Corridor: The current demonstration is limited to a single urban corridor. Future work will involve scaling the approach to larger and more interconnected regions.
  • Real-World Trial: The system is currently being trialled in the Kirklees region, with plans to expand testing to other areas in England. Further experimental analysis is ongoing.
  • Dynamic Network Changes: While the paper mentions that accidents or road works are handled by modifying link characteristics, a more robust mechanism for dynamically updating the network topology itself might be a future consideration.

Summary updated

BibTeX
@inproceedings{bhatnagar2023automated,
  author = {Saumya Bhatnagar and Rongge Guo and Keith McCabe and Thomas L. McCluskey and Francesco Percassi and Mauro Vallati},
  title = {{Automated Planning for Generating and Simulating Traffic Signal Strategies}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2023},
  doi = {10.24963/ijcai.2023/830},
}
Figure