Reproduce “AgentSUMO: An Agentic Framework for Interactive Simulation Scenario Generation in SUMO via Large Language Models” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Minwoo Jeong, Jeeyun Chang, Yoonjin Yoon

AgentSUMO is an agentic framework developed by researchers at KAIST that leverages large language models to enable non-expert users to interactively generate and analyze complex traffic simulation scenarios within SUMO. It demonstrates the capacity to translate abstract policy goals into detailed, executable simulations and provides interpretable results for urban planning and policy evaluation, outperforming previous LLM-SUMO integrations by offering a goal-oriented, interactive workflow.

Key insights

  • LLM-based agents can adapt their reasoning depth and interaction strategies to tasks ranging from simple commands to complex, multi-constraint problem-solving, enhancing both efficiency and analytical rigor.
  • Implementing a 'clarify-before-execute' principle with plan confirmation and parameter validation significantly improves transparency, user control, and the reproducibility of complex simulation studies.
  • A standardized Model Context Protocol (MCP) effectively decouples LLM reasoning from domain-specific tools, fostering a modular, extensible, and robust integration for complex simulation functionalities.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.