Proceedings of the 2009 Winter Simulation Conference M. D. Rossetti, R. R. Hill, B. Johansson, A. Dunkin and R. G. Ingalls, eds. TECHNIQUES FOR RAPID INITIALIZATION IN IN-VEHICLE TRAFFIC SIMULATORS Vishwanath Palagummi Michael Hunter Richard Fujimoto Computational Science and Engineering School of Civil and Environmental Engineering Georgia Institute of Technology Georgia Institute of Technology Atlanta, GA 30332, USA Atlanta, GA 30332, USA ABSTRACT On-line in-vehicle traffic simulation has been proposed as a means to provide predictions of future states of a traffic network based on current traffic conditions. The area covered by an in-vehicle simulation may change dynamically during the ve- hicle’s journey. This paper is concerned with the issue of initializing the state of new regions that are added to a running mi- croscopic traffic simulation. A warm-up period is required where the new road segments must be populated with simulated vehicles. Techniques that convert vehicle flow rate and queue length information to vehicle positions and speeds are pro- posed to minimize this warm-up period. These techniques are evaluated and compared with respect to estimated travel times of vehicles traveling through different paths in the road network under different traffic conditions. Travel time predictions are compared to a microscopic simulation of the entire region. 1 INTRODUCTION The last decade has seen dramatic advances in the development and deployment of technologies for sensors, mobile compu- ting, and wireless communications. Use of on-line simulation as a means to better manage and control operational systems has been receiving increased attention. This trend is particularly true in transportation systems where vehicles are now rou- tinely equipped with global positioning system (GPS) devices, route planning software, and on-line feeds of current traffic conditions derived from instrumented roadways. On-line simulations, fed by real-time data, can rapidly predict future system states for use in planning and management for individual vehicles, and potentially, for the system as a whole. Here, we are concerned with on-line simulations embedded in vehicles in order to predict future states of portions of the transportation network that are of interest to the traveler. We envision collections of mobile in-vehicle simulators, each modeling some portion of the road network. The road network may be viewed as being divided into a set of non-overlapping zones or grids, and each in-vehicle simulation selects certain grids to include in its model. For example, the vehicle may choose to model the next few grids in which it intends to travel on its current trip. As the vehicle moves, it may wish to add new grids into the area it is modeling, while deleting grids for areas in which it no longer has an interest, e.g., because it has completed travel through those regions. The addition of a new grid to the running simulation requires specification of the initial state for the new region. The required information in- cludes relatively static information (such as road topology) to dynamic, but somewhat predictable, information (such as sig- nal control) to rapidly changing, less predictable dynamic information (such as vehicle flow rate data or information concern- ing individual vehicles, i.e. exact position, current acceleration, etc). Here, we are concerned with this latter information, and specifically, determination of initial state information of microscopic traffic simulations based on current flow rate informa- tion derived from infrastructure sensors or through other means such as in-vehicle sensing. We assume topology, signal tim- ing and routing information is available from the infrastructure or historical data. The problem addressed here concerns the establishment of initial vehicle position and speed to populate a microscopic traffic simulation based on on-line data report- ing vehicle flow rates on segments of the region that is to be added. 2446 978-1-4244-5771-7/09/$26.00 ©2009 IEEE