Reproduce “Relational Multi-Agent Reinforcement Learning for Dynamic Pricing in High-Speed Railway Markets” 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

Enrique Adrian Villarrubia-Martin, David Muñoz-Valero, Luis Rodriguez-Benitez, Giovanni Montana, Luis Jimenez-Linares

Villarrubia-Martin et al. developed RACHE, a multi-agent reinforcement learning framework that employs an entity-based graph model for dynamic pricing in liberalized high-speed railway markets. The method achieved the highest total revenue in simulated scenarios of varying complexity, outperforming established baselines by modeling fine-grained service relations and enhancing relational reasoning.

Key insights

  • All three defined heterogeneous relation types (competition, connectivity, coordination) contribute positively to agent decision-making, with the competition relation demonstrating the most significant impact on performance.
  • A two-layer R-GCN configuration proved optimal for message passing, effectively capturing relational context without excessive 'smoothing' of node features.
  • Decoupling the representation learning from the actor's computational graph during training stabilized individual agent dynamics and led to more consistent revenue increments.

Summary by alphaXiv.

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