Quality Evaluation of Vehicle Navigation with Cyber Physical Systems Yang Yang * , Xu Li † , Wei Shu ♦ and Min-You Wu * * Department of Computer Science and Engineering, Shanghai Jiao Tong University, China † Department of Computer Science and Engineering, State University of New York at Buffalo, USA ♦ Electrical and Computer Engineering Department, The University of New Mexico, USA Abstract-In this article, we focus on a typical application of Cyber Physical System (CPS), i.e., vehicle route navigation. Two fundamental problems have been examined: 1) How different are various vehicle routing algorithms? 2) How valuable is real-time traffic information or historical traf- fic information in helping vehicle routing? Different from most previous works based on simulations, we presented performance comparisons of four routing algorithms using real GPS sensory data from 4000 taxis. It has been shown that using real-time traffic information could substantially improve the quality of vehicle path routing. This can bene- fit an individual driver, because through our evaluation results on realistic vehicle traces we found that the paths selected by taxi drivers are not as good as expected. More importantly, utilizing real-time information could improve global transportation efficiency in terms of dispers- ing/managing traffic, which plays a key role in construct- ing an effective vehicular CPS. Keywords-Vehicle path routing, real-time traffic informa- tion; Vehicle Infrastructure Integration; Cyber Physical System; IntelliDrive I. INTRODUCTION Recently, academic and industry community proposed the idea of communications and integration among vehicles and infrastructure within a transportation system, namely, Vehicle Infrastructure Integration (VII). The objective of this idea is not only to passively monitor traffic but also to actively con- trol and manage traffic, which is also called Cyber Physical Systems (CPS) in computer science domain. In this paper, we focus on a typical CPS application, i.e., vehicle route navigation [1][2], in which vehicles are assumed to be equipped with on-board navigation system. Different from most previous works, which mainly discussed algorithm design, we investigate the advantage of utilizing real-time traffic information for routing. We want to answer two fun- damental questions: 1) How different are various vehicle routing algorithms? 2) How valuable is real-time traffic in- formation or historical traffic information in helping vehicle routing? Since large-scale real deployment leads to considera- ble cost, most of existing works relied on simulations or artifi- cial traces, which have limitations to reflect realistic transpor- tation system and driver behaviors. Fortunately, we are able to utilize real-time data from GPS sensors of 4,000 taxis in Shanghai [6][8]. We mapped these data onto a digital map to trace vehicles and to evaluate various routing algorithms. We found that both historical and real-time traffic information can improve the performance of vehicle routing. This paper pro- vides a first-hand report about the worthiness of real-time traf- fic information for vehicle route navigation, which may serve as a guideline when deploying large-scale vehicular navigation systems. Some interesting insights have also been presented, e.g., we found that the paths selected by taxi drivers are not as good as expected. It is worth noting that our results not only showed how individual driver can benefit from real-time traf- fic information, but also it demonstrated that utilizing real- time information could improve global transportation efficien- cy in terms of dispersing/managing traffic, which contributes to a key component of an effective vehicular CPS. II. RELATED WORK Finding shortest path is a classical problem in graph theory. Most of related problems can be finally transformed to a shortest path problem, which has two properties: static or dy- namic, deterministic or stochastic [5][11]. For the static de- terministic case, all data are known in advance and they are independent on time [12] and this is a classical shortest path problem [2]. For the static stochastic case, the input data is both random and time-independent. Eiger et al [3] described that when the utility function is linear or exponential, an effi- cient Dijkstra-type algorithm can be used to find the optimal path. For the dynamic stochastic case, the input data is also random and time-dependent. Miller-Hooks [9] denoted that a single path cannot provide an adequate solution for a given source-destination pair at a specific departure time, because the optimal path depends on intermediate information. For the dynamic deterministic case, all data are also known in advance and they are time-dependent. Kaufman and Smith [7] showed that the standard shortest path (such as the Dijkstra [2] algo- rithm) can be used as long as the network satisfies a determi- nistic consistency condition (refer as FIFO). Chabini's work [1] showed the optimal routing path equations and the corres- ponding proofs in single source with ADET algorithms and single destination with SDOT algorithm. By using time space network, Pallottino [10] presented a general “chronological” algorithmic paradigm, called Chrono-SPT, which can be used to get the optimal routing path in discrete-time dynamic net- works. Overall, those existing works are based on specific assump- tions and simulations. Testing of various algorithms in the real traffic network is seldom to be found. We have built up a ve- hicular CPS, called the Shanghai Urban Vehicular Network (SUVnet) by collecting the GPS data from thousands of taxis. The SUVnet has been initially designed and developed by our research team at Shanghai Jiao Tong University. This platform has provided us with a foundation for real time information collection and dissemination, as reported in our previous work [6]. We have also investigated the traffic-monitoring perfor- mance obtained by SUVNet and showed that the estimations of traffic statuses are reasonably accurate [9]. Therefore, esti- mations of the traffic status have been used in this article for vehicle path routing. III. VEHICLE ROUTING ALGORITHMS We first introduce some terms and performance metrics used in this article. A traffic network G (N, E) includes a set of nodes (intersections) and a set of edges (road sections). Con- 978-1-4244-5637-6/10/$26.00 ©2010 IEEE This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE Globecom 2010 proceedings.