The MARL-OD-DA framework redesigns multi-agent reinforcement learning by using OD-pair–level agents and Dirichlet-based continuous routing actions, enabling scalable and stable traffic assignment in ...