Decoupled Intelligence: Multi-Agent LLM Framework Automates Traffic Simulation in SUMO

A new research paper introduces Decoupled Intelligence, a multi-agent LLM framework designed to automate the entire lifecycle of traffic simulation in SUMO (Simulation of Urban Mobility). Traditional monolithic LLM agents struggle with complex end-to-end simulation workflows, leading to reasoning failures, parameter inconsistency, and poor state management. Decoupled Intelligence solves this by breaking the pipeline into five specialized agents: Planner (designs the simulation), Builder (constructs the road network), Demand (generates traffic demand), Runner (executes the simulation), and Analyst (evaluates results). A high-level reasoning Orchestrator coordinates these agents, leveraging the Model Context Protocol (MCP) to maintain persistent state and ensure seamless data handover across distributed actions. This architecture enables a closed-loop refinement process where simulation outcomes are iteratively optimized to meet user-defined Key Performance Indicators (KPIs).

Experimental results through role ablation studies confirm that the multi-agent approach significantly outperforms single-agent baselines in task success rate and parameter accuracy. Case studies on real-world network extraction and traffic optimization further demonstrate the system's ability to bridge natural language intent with low-level simulation execution. By allowing researchers and urban planners to describe desired traffic scenarios in plain language and have the system autonomously generate, run, and refine simulations, Decoupled Intelligence represents a major step toward autonomous urban planning and intelligent transportation analysis. The paper is available on arXiv (2605.27685) and highlights a promising path for integrating LLMs with microscopic traffic simulation tools.