As transit agencies test out emerging technologies for providing service, they encounter a challenge of having limited data compared to conventional transit operations with well-designed survey processes and an existing rider base. In such circumstances, the selection of new routes need to play a double role, not just to provide access to travelers, but also to act as “sensors” to learn how the demand responds. We developed a framework for incrementally growing out such a transit network that uses these optimal learning techniques to decide where next to expand. His tests identify the circumstances when “knowledge gradient” policies with correlated beliefs can best be fashioned into such sequential transit network design processes — a next step toward AI-driven transit network evolution that can help expedite the adoption of new transit technologies and policies.

Paper can be found here: https://www.sciencedirect.com/science/article/pii/S1366554524002989

We developed this model framework to allow policymakers and mobility market regulators to evaluate different designs and scenarios: different mixes of mobility operators and their coverage areas, pricing policies, service capacities, subsidy provision, etc., considering both fixed route and on-demand service providers. As a generalized tool, it can be used to examine special cases: mobility hub design, electric MaaS markets, first/last mile design for transit providers, cooperation and competition among mobility providers, among others. The code to the model solution algorithm can be found on Github (https://github.com/BUILTNYU/MaaS-Platform-Equilibrium).

Paper can be found here: https://www.sciencedirect.com/science/article/abs/pii/S019126152400184X

Public data for transportation planning often undersamples underserved population groups. For example, the margin of error (MOE) for population counts at census tract level for people with disabilities can make up 36% of the estimate. If we are to move towards a more equity-aware profession, the data we use need to account for this. In this work with Bingqing Liu and Farnoosh Namdarpour, we develop an algorithm to aggregate zones together to ensure sufficient data reliability for different underserved groups. We apply this algorithm for NYC to produce a set of 574 zones that result in lower MOEs compared to similarly aggregated zonal structures. We also show that this reduction in error translates to reduced variance in downstream transportation applications, such as population synthesis. The data and algorithm are publicly available and can work well with data standards like MDS.