AI-Driven Urban Mobility Optimisation

Most traffic-AI is validated under conditions cities can't reproduce — clean feeds, perfect connectivity, cooperative vehicles. This project starts from the opposite premise: real infrastructure is noisy, intermittent and unforgiving. At its core is a hybrid controller that pairs learned models with a deterministic, rule-based fallback — when the model hesitates or sees data it doesn't trust, control degrades gracefully to a transparent baseline. Funded by an ARC Linkage grant with the City of Brisbane, it has run for over fourteen months without a safety incident, cutting intersection delay while holding a hard 800-millisecond decision ceiling.