AlphaTransit: Learning to Design City-scale Transit Routes

Researchers introduced AlphaTransit, a search-based bus network design system that combines Monte Carlo Tree Search with a neural policy-value network.

Unlike rollout-based methods, it selects route extensions directly, helping it handle delayed feedback and route interactions in transit planning.

On a Bloomington benchmark with realistic demand and road data, AlphaTransit achieved the best service rates in both mixed and full demand settings.

It outperformed reinforcement learning and unguided MCTS, suggesting learned guidance plus search can improve city-scale transit design.