This idea will explore traffic signal control done by AI agents that negotiate in natural language: a vehicle requesting priority, a traffic light balancing competing demands, a neighbouring intersection protecting its own green-wave timing. This project builds the foundational system: a working prototype combining a traffic simulator (SUMO) with large language model agents that negotiate over signal priority in realistic scenarios. You’ll design the agents, run and log their negotiations across dozens of scenarios, and systematically analyse where and why these negotiations break down — using a failure-mode taxonomy from recent multi-agent AI research to categorise the problems you find.
Reference:
Mert Cemri, Melissa Z. Pan, Shuyi Yang, Lakshya A. Agrawal, Bhavya Chopra, Rishabh Tiwari, Kurt Keutzer, Aditya Parameswaran, Dan Klein, Kannan Ramchandran, Matei Zaharia, Joseph E. Gonzalez, and Ion Stoica. “Why Do Multi-Agent LLM Systems Fail?” 39th Conference on Neural Information Processing Systems (NeurIPS 2025), Datasets and Benchmarks Track.