CIVIQ· A Hierarchical Multi-Agent Coordination FrameworkSolving Urban Congestion Through Hierarchical Coordination.
An undergraduate thesis applying Hierarchical Multi-Agent Reinforcement Learning to urban traffic management. Vehicles act as intelligent agents that cooperatively learn routing decisions — reducing congestion across a simulated road network.
Simulation Controls
Redefining Urban Traffic Management
CiViQ is a smart traffic coordination system tested on a 2 km² simulation of BGC. It works on three levels: individual vehicles, 17 roadside units that coordinate nearby intersections, and a central server that oversees the whole network. Across three traffic volumes — light, moderate, and heavy — CiViQ moves 37.4% more vehicles per hour than the single-AI baseline and cuts average wait times by 18.8% when roads are at their busiest.
Scalable by Design, Fast in Practice
Simulated on a 2 km² model of BGC, CiViQ was tested at three traffic levels. When roads are at peak load (2,000 vehicles/hr), it clears 2,273 vehicles per hour — 37.4% more than the single-AI approach — and cuts average wait time from 23.75 sec to 19.29 sec. At moderate traffic, the single-AI approach is actually 11.3% faster in travel time (108.3 vs. 122.1 sec), showing that CiViQ's coordination pays off most when the network is under pressure. Fuel efficiency also improves: 21.2 L/100km for CiViQ vs. 23.4 L/100km — a 9.4% reduction.
The Researchers
