Indian researchers' SIGMA AI slashes traffic wait times 40%
New AI traffic system reduces emergency vehicle wait times by 40% with LLM-guided control
SIGMA is a new reinforcement learning framework for traffic signal control that uses a large language model to translate natural-language emergency commands into adaptive priority settings—no manual reward engineering required. It also learns orientation-invariant features so it can transfer across geometrically similar four-way intersections, with offline-to-online training for stable real-world adaptation. Tested in SUMO simulations across four Kolkata-based urban intersections against fixed-time, actuated, and DQN controllers, SIGMA reduces average and emergency waiting times, shortens queue lengths, and boosts throughput, with statistical reliability checks for graceful degradation and emergency service levels.
- SIGMA uses RL + LLM to dynamically optimize traffic signals across multiple objectives (throughput, fairness, emergency priority) without manual reward tuning
- Reduces average wait times by 40%, emergency response times by 40%, and increases throughput by 15% in Kolkata simulations
- Features rotational augmentation for geometric transferability and offline-to-online learning for stable deployment
Why It Matters
Could revolutionize smart city infrastructure by cutting emergency response times while reducing urban congestion through AI-optimized traffic control