Reproduce “Joint-Local Grounded Action Transformation for Sim-to-Real Transfer in Multi-Agent Traffic Control” 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
Justin Turnau, Longchao Da, Khoa Vo, Ferdous Al Rafi, Shreyas Bachiraju, Tiejin Chen, Hua Wei
Researchers from Arizona State University developed Joint-Local Grounded Action Transformation (JL-GAT), a framework that reduces the sim-to-real gap in multi-agent reinforcement learning for traffic signal control by incorporating local inter-agent dynamics. This approach significantly lowered performance discrepancies in simulated adverse weather across different network sizes, improving traffic flow metrics.
Key insights
- Leveraging local joint information from neighboring agents within a decentralized grounding framework effectively balances scalability with the ability to capture important multi-agent dynamics.
- The "cascading invalidation effect" is a unique challenge in multi-agent grounding, demonstrating the necessity of specific mitigation strategies for stable training.
- Ablation studies confirm that incorporating both neighboring states and actions into the forward and inverse models is critical for JL-GAT's ability to effectively bridge the sim-to-real gap.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.