Reproduce “Prompt to Transfer: Sim-to-Real Transfer for Traffic Signal Control with Prompt Learning” 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
Longchao Da, Minquan Gao, Hao Mei, Hua Wei
Arizona State University researchers introduced PromptGAT, a method that leverages Large Language Models (LLMs) to enhance sim-to-real transfer for Reinforcement Learning policies in Traffic Signal Control. The approach consistently reduces the performance gap across varied real-world conditions by incorporating LLM-inferred traffic dynamics.
Key insights
- LLMs can provide rational and consistent quantitative inferences about physical system dynamics when given contextual information, aligning with human understanding of traffic behavior.
- Incorporating LLM-derived domain knowledge significantly improves the accuracy of the forward model within the GAT framework, allowing it to better approximate real-world transitions.
- The improved accuracy of the forward model directly correlates with a greater reduction in the sim-to-real performance gap for RL policies in traffic signal control.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.