SimuAgent: An LLM-Based Simulink Modeling Assistant Enhanced with Reinforcement Learning

While LLMs excel in text-based code automation, their potential in graph-oriented engineering workflows like Simulink remains underexplored. SimuAgent is an LLM-powered modeling and simulation agent for Simulink, which replaces verbose XML with a concise Python representation, improving interpretability and enabling fast, in-process simulation. It employs a two-stage plan-execute architecture and introduces Reflection-GRPO (ReGRPO) for sparse rewards in long-horizon tasks. Experiments on SimuBench demonstrate SimuAgent's faster convergence and higher modeling accuracy compared to standard RL baselines and even GPT-4o. It offers a privacy-preserving, cost-effective solution for industrial model-driven engineering.