All projects2 / 6
Team project
traffiQ
AI agents that learn to run the traffic lights of a simulated city.
- Year
- 2026
- Role
- Simulation & AI integration
- Context
- Team project
Overview
TraffIQ is an intelligent traffic simulation project using SUMO and a reinforcement learning agent. The objective was to integrate a DQN model capable of controlling traffic lights across different traffic scenarios. I worked on the simulation environment, AI agent integration, and multi-agent architecture. The system combines neural network decisions with a route-aware safety filter and a Deadlock Guard to prevent gridlock situations while improving overall traffic flow.
Built with
Python- SUMO
- Reinforcement learning (DQN)
Collaborators
What I did
- 1Development of the traffic simulation environment using SUMO
- 2Integration of a DQN model based on reinforcement learning
- 3Implementation of a multi-agent architecture with one agent per intersection
- 4Development of the observation and decision pipeline using 50×50 RGB images
- 5Optimization of neural network calls using AgentCallManager
- 6Implementation of a route-aware safety filter for conflict resolution
- 7Development of a Deadlock Guard to prevent gridlock situations
- 8Automatic generation of road networks and traffic scenarios
- 9Implementation of reward strategies based on waiting time, congestion, and arrived vehicles
- 10Development and validation of multi-intersection scenarios
Walkthrough
A quick look at the project in use.
Gallery
