Onabanjomicheal/Adaptive-Traffic-Signal-DQN

🚦 Adaptive traffic signal control using Deep Q-Networks (DQN) in SUMO — optimized for Lagos, Nigeria # 🚦 Lagos Traffic Optimization with Deep Q-Networks (DQN) in SUMO A Reinforcement Learning model that clears traffic in Lagos 59% faster than traditional systems—tested on real maps using SUMO and Deep Q-Networks. ## Why This Matters 🔥 Lagos traffic isn’t just inconvenient—it’s catastrophic. The Oke Afa Roundabout and Isolo–Mushin Corridor are among the city's most congested intersections. Traditional fixed-timing systems collapse under pressure. This project uses Deep Reinforcement Learning to fix that. Our DQN-based signal controller outperforms traditional systems, cutting travel time by over 30%, increasing speed by 50%, and completing nearly 2x more trips—all in a realistic SUMO simulation built from actual OSM map data. ## Table of Contents - Project Goals - Agents Built - How It Was Tested - Performance Highlights - Demo Video - Stack and Tools - Contributing - What to Do Next - Author ## Project Goals 🎯 To simulate real-world traffic conditions at multiple intersections, including the Oke Afa Roundabout and adjacent junctions. - Simulate Lagos traffic at key intersections (Oke Afa Roundabout, Isolo–Mushin) using SUMO. - Train a Deep Q-Network agent to dynamically adapt traffic signals in real-time. - Benchmark against Fixed-Time and classical Q-Learning agents. - Measure speed, throughput, emissions, and travel efficiency. ## Agents Built 🧠 | Agent | Description | | ---------------------- | -------------------------------------------------------------------------------- | | ✅ **DQN Agent** | Learns traffic light timing via neural networks. Maximizes flow, minimizes wait. | | ⬜ **Q-Learning Agent** | Table-based RL. Simple but effective baseline. | | ❌ **Fixed-Time Agent** | Static signal schedule. Old-school, non-adaptive benchmark. | ## How It Was Tested 🧪 - SUMO simulation (command-line, 0.5s s …