We looked at data from Portland, Austin, Chicago, Hoboken, & NYC to derive a forecast model for e-scooters (that is pegged to fleet size) and use it to explain the trips that could be substituted from different modes, including access trips to public transit, by distance. For a fleet size of 2000 e-scooters in Manhattan, we predict 75K daily e-scooter trips that could translate to $77M annual revenue. We find that the distance structure of revenue from transit access/egress trips differs from that of other substituted trips. Given e-scooter pilots (e.g. Lime) in NYC now, our insights could be timely for cities to anticipate how e-scooters will impact the mobility ecosystem. The research was partially funded by C2SMART and the paper can be found here:

https://www.sciencedirect.com/science/article/abs/pii/S1361920921001930

We recalibrated our synthetic population and MATSim-NYC (see https://www.sciencedirect.com/science/article/abs/pii/S0967070X20309483) to fit COVID work-from-home conditions in 2020 and used it to evaluate varying social distancing requirements from the transit service. The study showed that full reopening may result in higher traffic levels than pre-COVID setting because of adverse impact on travelers’ preference for shared use transportation modes. The study suggests more attention to be paid toward managing transit capacity, traffic in Manhattan, and micromobility provision as it will play an important role.

Funding support from C2SMART/USDOT #69A3551747124; the paper can be found here.

We generalized Wilson’s classic gravity model to a multilayer one that jointly distributes passenger ODs as well as mobility service vehicle ODs. This was tested on Chicago Uber data and shown to distinguish between shared and solo rides. The work is meant for planning agencies working with limited, privacy-protected data from fleet-based mobility companies.

The work was funded by C2SMART (USDOT #69A3551747124) and NSF CMMI-1652735. Link to paper here.

Transportation network inference can be statistically costly, but if we can reverse engineer network congestion levels to fit route observations, that can produce a statistically cheaper approach to monitor traffic in cities. This method was retroactively tested assuming a network monitoring system based on taxi data from Wuhan as mobile sensors, showing that even increasing monitoring of just one OD pair’s routes to two OD pairs can improve the correlation coefficient between predicted and observed from 0.23 to 0.56. Should be of interest to city monitoring agencies and companies like Google, Uber, and DiDi.

Funding from NSF CMMI-1652735 with collaboration from Dr. Xintao Liu at Hong Kong Polytechnic University. The paper can be found here. 

This collaborative work with Prof. Ozbay’s group, and consists of a 2-year grant from C2SMART, is now published in Transport Policy. The paper details the calibration and validation process for the MATSim-NYC and applied to the city’s congestion pricing plans. The findings suggest that there is a pricing peak over which the benefits to Manhattan come at the expense to the residents of the other boroughs, and suggest a redistribution of revenues to transit services in those boroughs.

The paper can be found here.

This paper is a part of a multi-year project in developing a citywide virtual test bed for NYC; details of this broader project can be found here: https://www.linkedin.com/pulse/matsim-nyc-virtual-test-bed-policies-transportation-nyc-joseph-chow/

The data can be found here (updated 12/12/22):

https://zenodo.org/record/7430184

Like physical internet search engines, can we integrate destination recommendation with on-demand mobility service (e.g. call an Uber, and let it recommend a restaurant to you and deliver you there)? Reinforcement learning algorithms in this context can be inefficient due to the added context from the routing service, but we identify more amenable conditions for operating such a feature, which might help seniors with less real-time information access, or post-COVID with more uncertainty on destination information. This work funded by C2SMART and NSF and should be of interest to TNCs or microtransit services.

The full paper is available here.

For multi-agent simulations, it is important to have a well-calibrated underlying synthetic population. As a precursor to the MATSim-NYC model, we created a synthetic population for NYC of 8M+ individuals for the year 2016. The population includes use of mobility options like Citi Bike and ridehail services. The dataset is publicly accessible (https://dw.tandoncsmart.com/dataset/synthetic-population) and we illustrated its use in evaluating the prior Amazon LIC HQ plan as well as Citi Bike’s expansion plan. The work was a joint 2-year project with Professor Ozbay funded by C2SMART with support from the NYC DOT.

Link available here: https://www.sciencedirect.com/science/article/pii/S096585642030745X