TNC Assignment

When a TNC trip request is made, the fleet operator attempts to assign it to the closest available vehicle to reduce total empty vehicles miles traveled (eVMT), as well as to minimize the traveler waiting time. Assignment is strategy-specific with the default being a zone-based passenger-vehicle assignment. The coordinate-based search utilizes vehicle coordinates, which are updated continuously throughout the simulation to create an R-tree. An R-tree is a graph-based pre-compiled library that provides computational advantages for graph-based operations. The R-tree improves spatial searches, and, therefore, the coordinates of an incoming request can be used to find the closest vehicle assessed using the Euclidian distance and the R-tree structure. Despite R-tree being an efficient spatial data structure, it may have some drawbacks. The Euclidian distance may not necessarily reflect the travel times between vehicles and requests since mean speeds vary along different routes in the network depending on congestion and network characteristics. Another aspect critical in large-scale simulations is the computational burden. R-tree’s insertions, deletions, and queries take considerably more time compared to basic data structures (on the order of milliseconds compared to microseconds for the latter). However, this does not significantly grow with fleet size for individual operations. While this is not an issue to simulate present-day TNC systems that enjoy a relatively small share of trips, it may be significant for future scenarios in which millions of trips are potentially served by SAV fleets.

In order to account for this, an adapted version (Gurumurthy et al., 2020) of the zone-based structure in (Bischoff and Maciejewski, 2016) is implemented to be able to simulate large-scale scenarios with a reasonably-low computation time. The zone-based architecture is generated using traffic analysis zones (TAZs) which is included in POLARIS. The TNC operator constructs an array of all neighboring zones, for each zone in the region, in ascending order of free-flow travel times defined with respect to a reference zone. This array is truncated using a predefined threshold for wait times so that a minimum LOS is maintained by the TNC service. By definition, the first zone in each array is the reference zone itself, which would mean minimal wait time. When a trip is requested, the operator checks for any available vehicle, starting from the origin zone and in the same order defined by the array, and assigns an unoccupied vehicle from the first neighboring zone that has a vehicle available. Since vehicles in these arrays are stored with longest-idling vehicle first, they need not be the closest available vehicle, but are expected to serve the request in reasonably less time. Both the assignment strategies described here can be adapted to match requests to current trips to simulate pooling. Maximum user wait time is a threshold the modeler can control. It refers to the maximum wait time for a vehicle once a TNC is assigned to a person. If there is no TNC within travel distance equivalent to the user’s maximum wait time, they cannot choose a TNC mode. The user can try choosing a different mode that maximizes their utility or try the TNC mode again, so long as vehicle assignment does not exceed one quarter of the maximum wait time. In effect, the total possible maximum user wait time is 1.25 times the threshold the modeler inputs.