Fairness in Transport Network Design - A Multi-Objective Reinforcement Learning Approach

Optimizing urban transportation networks can improve the lives
of millions of citizens worldwide. The problem of generating new
transportation lines, which maximize the levels of satisfied travel
demand is, however, a complex endeavor. This problem is known as
the Transport Network Design Problem (TNDP) and it is NP-hard.
On top of efficiency concerns, it is nowadays fundamental to also
consider the development of transportation systems that contribute
to alleviating social inequalities. Which technical approaches can
we employ to tackle both efficiency and fairness in TNDP? In this
paper, we explore Multi-Objective Reinforcement Learning (MORL)
as a tool to design efficient and fair transportation networks. We
start by formulating Multi-Objective transport network design problems
as Multi-objective Markov decision processes. We highlight
the main challenges of introducing multiple objectives in TNDP.
Finally, we describe novel methodologies that can be used to tackle
this problem. With this paper, we hope to start a line of research
that can provide suitable decision support for TNDP, by providing
alternative solutions with different trade-offs between (different
metrics of) fairness, efficiency, and cost.