Transiting to fair cities Reinforcement learning and multi-agent systems for equitable transport network design

As humanity becomes increasingly urban, cities have emerged as both engines of progress and sites of inequality. Urban transport systems are central to this transformation: they link people to jobs, education, and opportunity, yet they can also reinforce social disparities. This thesis explores how artificial intelligence (AI), particularly reinforcement learning (RL) and multi-agent modeling, can support more equitable, efficient, and sustainable transport planning.
The thesis is structured into three parts: (I) the agent as a transport planner, (II) the agent as a commuter, and (II) the agent as a planning facilitator. In the first part, transport network design is formulated as an optimization problem in which RL agents generate transport lines under fairness constraints. Novel tabular and multi-objective RL methods are proposed to balance efficiency and equity, achieving performance comparable or better than deep RL baselines while remaining interpretable and computationally efficient.
The second part models commuters as adaptive agents to study how transport interventions influence school segregation and congestion. Using agent-based and game-theoretic models, we show that targeted network improvements can reduce segregation, but that disparities in commuters’ learning rates may also reproduce persistent inequality in mobility outcomes.
The third part introduces a conceptual framework that integrates community feedback into the planning process. Rather than viewing AI as a purely top-down optimizer, this framework positions it as a collaborative tool that bridges technology, policy, and society.
Together, these contributions demonstrate how AI can advance transport planning that promotes not only efficient mobility, but also fairness, inclusivity, and sustainability.