Abstract
As public transport authorities transition from pilot to scaling on-demand shared autonomous vehicle (SAV) services, they face a persistent challenge—improving economic viability without undermining equitable and affordable access. Dynamic pricing offers a potential mechanism to manage this trade-off by aligning fares with real-time demand and capacity conditions. However, its application in public SAV systems requires modelling and solution approaches that reflect public-service objectives. This paper develops a deep reinforcement learning framework to study dynamic pricing in a publicly operated SAV service. The pricing problem is formulated as a continuous-action Markov Decision Process and solved using Proximal Policy Optimisation, with capped fare adjustments. Using a simulation calibrated with operational data from an on-demand SAV pilot in Oslo, we compare static pricing with capped and uncapped dynamic pricing policies. The results indicate that static pricing preserves accessibility but constrains economic performance, while unconstrained dynamic pricing improves profitability at the cost of increased price dispersion and reduced service accessibility. In contrast, a capped dynamic pricing policy achieves a more balanced outcome by maintaining accessibility and price stability while retaining most of the efficiency gains from adaptive pricing. These findings suggest that dynamic pricing can support the economic viability of public SAV services when implemented as a regulated policy instrument. The proposed framework provides a policy-relevant basis for analysing efficiency–equity trade-offs in dynamic pricing for public MaaS services.

This publication has 19 references indexed in Scilit: