Developing and deploying machine learning systems for real-time adaptive signal control and network-wide flow optimisation in mid-sized Australian cities.
Overview
This project develops a suite of AI-powered tools for urban traffic management that are explicitly designed for deployment constraints — intermittent connectivity, heterogeneous sensor infrastructure, and real-time latency requirements — rather than for idealised laboratory conditions.
The core contribution is a hybrid architecture that couples learned predictive models with rule-based fallback systems, ensuring safe and predictable behaviour even when the AI component degrades.
Research Questions
- What model architectures achieve the best latency-accuracy trade-off for intersection-level decision support?
- How can transfer learning reduce the data requirements when adapting models to new city contexts?
- What governance frameworks should accompany AI deployment in safety-critical transport infrastructure?
Methodology
The project operates across three phases:
Phase 1 (2023): Baseline data collection and model development across four instrumented intersections in Brisbane.
Phase 2 (2024–2025): Controlled deployment with human-in-the-loop oversight; iterative model refinement based on operational logs.
Phase 3 (2026): Evaluation, generalisation study, and policy brief preparation for Department of Transport.
Outcomes to Date
- Two peer-reviewed journal articles (J1, J2 in publications list)
- One best-paper award at IEEE ITSC 2025
- System deployed and operating for 14 months without safety incident
- Dataset of 18 months of intersection telemetry prepared for open release
Industry Partners
City of Brisbane, Transurban, Q-Traffic Systems Pty Ltd.
Prof. Sarah Chen
Massachusetts Institute of Technology
Dr. James Okonkwo
University College London
Dr. Kenji Tanaka
Kyoto University