Article Transportation Research Record 2018, Vol. 2672(47) 219–230 Ó National Academy of Sciences: Transportation Research Board 2018 Article reuse guidelines: sagepub.com/journals-permissions DOI: 10.1177/0361198118796402 journals.sagepub.com/home/trr Time-Dependent Intermodal A* Algorithm: Methodology and Implementation on a Large-Scale Network O ¨ mer Verbas 1 , Joshua Auld 1 , Hubert Ley 1 , Randy Weimer 1 , and Shon Driscoll 1 Abstract This paper proposes a time-dependent intermodal A* (TDIMA*) algorithm. The algorithm works on a multimodal network with transit, walking, and vehicular network links, and finds paths for the three major modes (transit, walking, driving) and any feasible combination thereof (e.g., park-and-ride). Turn penalties on the vehicular network and progressive transfer penal- ties on the transit network are considered for improved realism. Moreover, upper bounds to prevent excessive waiting and walking are introduced, as well as an upper bound on driving for the park-and-ride (PNR) mode. The algorithm is validated on the large-scale Chicago Regional network using real-world trips against the Google Directions API and the Regional Transit Authority router. Transportation network companies (TNC) such as Uber and Lyft, car and bike sharing companies, on-demand transit services, the forthcoming connected and autono- mous vehicle (CAV) technologies, as well as the increas- ing availability of real-time traffic and transit information, give travelers the opportunity to evaluate their multiple routing options and make better-informed decisions. The advent of real-time control and manage- ment technologies, and vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communication technolo- gies provide opportunities to increase mobility, accessi- bility, throughput, and safety in the entire transportation network. These advancements call for a comprehensive modeling of the transportation system as an integrated multimodal network. Most of the existing transportation network modeling literature focuses either on the vehicular traffic network, or the transit network. The full integration of the two major modes is usually limited to small hypothetical net- works, which is not practical for large cities. At the large scale, the integration is performed in an ad-hoc fashion by which separate models communicate with each other at designated outer iterations. The drawbacks of this approach can be listed as follows:  The interaction between the transit traffic and vehicular traffic cannot be modeled properly. For instance, transit buses share the street network with passenger cars and affect the performance characteristics of each other.  The modeling of intermodal routing such as park- and-ride (PNR), kiss-and-ride (KNR), taxi/TNC/ CAV to transit, and taxi/TNC/CAV after transit is limited.  The modeling of en-route mode switching is limited.  As a result of all of the above, the integration of the transportation supply model with the activity- based demand models (ABM) is limited. This paper proposes a flexible intermodal routing algorithm that can provide time-dependent shortest paths for conventional modes such as passenger car and walk- to-transit, as well as any feasible intermodal combination such as PNR, KNR, taxi/TNC/CAV before/after transit, and so on. The rest of the paper is structured as follows: The fol- lowing section provides a brief literature review on the shortest path problem and its multiple branches, 1 Argonne National Laboratory, Lemont, IL Corresponding Author: Address correspondence to O ¨ mer Verbas: omer@anl.gov