Answers are grounded in this paper and cite their sources.
Urban public transport fleets face challenges in balancing operational efficiency and passenger-related service quality. This systematic review, reported with reference to PRISMA 2020, synthesises 41 studies identified through five academic databases and search platforms between January 2018 and May 2026. The corpus encompasses direct public-transport applications, general Multi-Agent Reinforcement Learning (MARL) and Multi-Objective Reinforcement Learning (MORL) research, and supporting literature from adjacent domains. The review examines MARL architectures, multi-objective reward formulations, coordination mechanisms, and operational outcomes while distinguishing application-specific findings from methodological insights. A conceptual Multi-Objective Partially Observable Stochastic Game (MO-POSG) formulation illustrates interactions between fleet-level objectives and passenger-related considerations. The synthesis discusses Centralised Training and Decentralised Execution (CTDE), value-based and policy-based learning, and scalarisation, constraint-based, and Pareto-based optimisation without assuming universal algorithmic superiority. Evidence from direct public-transport applications indicates potential benefits for coordinated vehicle control and service regularity, although reported passenger waiting-time, travel-time, and delay outcomes remain difficult to compare across studies. Offline MARL and conservative policy rectification are considered as potential approaches for limiting unsupported policy updates, without guaranteeing operational safety. Key deployment challenges include sensing uncertainty, adversarial vulnerabilities, and driver acceptance. Overall, the findings highlight limited application-specific validation and the need for transparent, comparable evaluation before broader real-world deployment.
No takes yet. Share an insight, caveat, or question.
Phan Gia Bao Le (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: