Investigators:
Dr. Michael Hunter, Associate Professor, Georgia Tech
Dr. Richard Fujimoto, Professor, Georgia Tech
Dr. Christos Alexopoulos, Associate Professor, Georgia Tech
Dr. Randall Guensler, Professor, Georgia Tech
Ya-Lin Huang, GRA, Georgia Tech
Project Overview:
Transportation is the largest industry in the world. Our transportation system significantly impacts every individual and the welfare of our entire nation in terms of economics, health, and quality of life to name a few. However, for many decades improvements to our ability to actively manage our surface transportation system in real-time have been stagnant. Today, wide-spread deployment of sensors, computers, and communications in vehicles and roadways is creating new challenges and opportunities to effectively exploit the wealth of real-time data and information that are becoming available. We attempt to capitalize on these rapid technology and communications advancements using ad hoc distributed simulations that features dynamic collections of autonomous simulations interacting with each other and with real-time data in a continuously running, real-time, distributed simulation environment. One can envision a distributed, adaptive, self-optimizing transportation infra-structure that can automatically reconfigure itself to maximize efficiency and minimize the effects of unexpected events ranging from everyday crashes to catastrophic natural or human generated disasters.
Our current research efforts are aimed at developing an embedded, distributed, simulation based transportation management system, combining in-vehicle simulators with information servers and simulations running within the roadside infrastructure. In our current implementation each participating vehicle contains a simulator that models the roadway network in the immediate vicinity of the vehicle, illustrated in Figure 1. Thus, as the vehicle traverses the network it will be simulating a dynamic roadway topology that must be continuously updated to reflect the current vehicle position.
The initial approach to implement an ad hoc distributed simulation system is a client/server architecture, where global state objects (i.e. simulation results from participating vehicles and sensor data) and associated composition functions (i.e. methods to aggregate data from the distributed simulations) are implemented at the server level (Fujimoto et al. 2007, Hunter et al. 2009). The server receives and stores state updates, compute composite values, and disseminates this information to other simulators. The key elements of our initial approach include Space Time Memory, State Aggregation, and Rollback Based Synchronization.
Our preliminary work assumed that the vehicle simulators were already configured to model the designated scenario, and focused on predicting the effects of changes in traffic patterns. One of our initial experiments involved a Manhattan-style 10 x 10 grid with two-lane, two-way roads. Forty participating vehicles (i.e. vehicles with on-board simulators) are distributed over the network at initialization, each simulating a group of intersections and connected roads, with the arrival rate on all boundary input roads initialized to a specified value. Each client simulation area depends upon the vehicle location and the direction in which the vehicle is traveling. The simulation region for each client is defined as the 5 x 3 intersection area in front of the vehicle assigned to it. If the network area in front of the client is smaller than the region, then the client only simulates the smaller extent.
One of experiments that were conducted to examine the feasibility of this approach involved an instance when the input arrival rate changes suddenly for the study network. Unlike the results from a steady state study, which experienced no rollbacks at the low input rate
settings, a distributed simulation will require rollbacks in order to successfully simulate the increase in traffic. Figure 2 focuses on a typical road segment along the east bound main artery. The set of experiments at each input rate is represented in the figure by three lines showing the average input rate over time of the replicated trials, the ad hoc simulation, and a single ad hoc distributed client (which illustrates how individual clients are reacting to the change in the arrival rate). The ad hoc simulation approach is able to track the unexpected change in input rate. Measurements on a variety of points in the network demonstrated that the ad hoc approach was able to track this change in traffic flow at different locations throughout the network.
Initial results support the potential of this approach. The above demonstrated that real-time transportation system monitoring and control is in fact feasible and will subsequently increase the system efficiency as facility managers will in-turn make better-informed decisions when operating transportation facilities.