A traffic light that sees the traffic
Ten years of algorithms and real deployments of smart traffic lights: what works in a real city and what still only exists in simulation.
The problem
Congestion in cities costs time, fuel and health. Fixed-cycle traffic lights do not react to real traffic, and over the last decade many approaches to smart control have appeared. They develop in different directions, though, and are rarely compared by the same criteria.
What we did
We reviewed the algorithms and real deployments of smart traffic lights from 2015 to 2025 and arranged them into a taxonomy:
- centralised optimisation;
- decentralised heuristics;
- max-pressure;
- model predictive control;
- learning-based methods, with a focus on single-agent and multi-agent reinforcement learning.
We also looked at the infrastructure that makes them possible: edge computing, on-device detection and vehicle-to-infrastructure communication (V2X). We compared how neighbouring intersections coordinate, how what was learned in simulation transfers to a real city, and how several goals are balanced at once: delay, emissions, safety, and priority for public transport and emergency services.
What it showed
We gathered the results of pilot projects and city platforms and classified the approaches by data, optimisation strategy and deployment maturity. This also shows the gaps: generalisation to new conditions, robustness under rare events and explainability of decisions. We close with directions for hybrid edge-cloud architectures, standardised interfaces and reproducible evaluation.
Why now
Many cities already have cameras, sensors and connected controllers, but do not know which approach to choose. The review gives them a map to find their way and shows what has been proven in real conditions and what still only exists in simulation.
The publication
Ivo Gergov, Rumen Doynov. Traffic Optimization with Smart Traffic Lights: An Overview of Algorithms and Implementations. Proceedings of the Technical University of Sofia 76 (2026). doi:10.47978/tus.2026.76.03.012