Multi-Agent System for Indian Highway Traffic Optimization

Design a multi-agent reinforcement learning (MARL) system that controls traffic signals across a 12-intersection grid of an Indian city (modeled in SUMO). Each agent learns to minimize local + neighbor delay. Compare independent learners (IQL) vs cooperative (QMIX). Quantify reduction in average delay vs fixed timing.

Course Learning Outcomes (CLOs):
CLO1: Apply MDP formulation to a real-world control problem.
CLO2: Implement single-agent + multi-agent RL algorithms.
CLO3: Engineer reward functions that balance global + local objectives.
CLO4: Evaluate RL policies using domain metrics (delay, throughput, emissions).
CLO5: Communicate results in IEEE-format research paper.

Industry/societal relevance: Indian Smart Cities Mission ($30B+ deployed) is procuring adaptive-signal systems; FAANG-equivalent prep for autonomy / RL roles at SARVAM, Bosch India, Ola Krutrim.