field. From multiagent systems and agent structure to ways of negotiating Detector Detector Detector between agents to control agent strat- egies, all these fields have had varying Trafﬁc management systems in degrees of success. Mainframe (1940s–1950s) centralized model (1960s) Now, the IT industry has ushered in the fifth computing paradigm: cloud computing. Based on the In- ternet, cloud computing provides on- demand computing capacity to indi- Detector Trafﬁc management systems viduals and businesses in the form of PC (1960s) with single points (1970s–1980s) heterogeneous and autonomous ser- vices. With cloud computing, users do not need to understand the details Organizer of the infrastructure in the “clouds;” they need only know what resources Coordinator Coordinator Coordinator they need and how to obtain appro- priate services, which shields the computational complexity of provid- Executor Executor Executor Executor Executor Executor ing the required services. In recent years, the research and Trafﬁc management systems application of parallel transportation LAN (1970s) in distributed model (1980s–1990s) management systems (PtMS), which consists of artificial systems, com- Mobile agent putational experiments, and parallel execution, has become a hot spot in the traffic research field.2,3 Here, the Internet term parallel describes the parallel interaction between an actual trans- portation system and one or more of its corresponding artificial or virtual Trafﬁc management systems counterparts.4 Internet (1980s–1990s) based on agents (2000s) Such complex systems make it diffi- cult or even impossible to build accu- Mobile agent rate models and perform experiments, ATS so PtMSs use artificial transportation systems (ATS) to compensate for this defect. Moreover, ATSs also help op- timize and evaluate large amounts of traffic-control strategies. Cloud com- Actual puting caters to the idea of “local transportation systems simple, remote complex” in parallel Cloud computing (2000s) Parallel transportation management systems (2010s) traffic systems. Such systems can take Figure 1. Relationship between the shifts in the computing and traffic management advantage of cloud computing to or- paradigms. ganize computing experiments, test the performance of different traffic agents were proposed to handle this to improve performance by reduc- strategies, and so on. Thus, only the vexing problem. Only requiring a ing communication time and costs. optimum traffic strategies will be used runtime environment, mobile agents This computing paradigm soon drew in urban-traffic control and manage- can run computations near data much attention in the transportation ment systems. This helps enhance74 www.computer.org/intelligent IEEE INTELLIGENT SYSTEMS
urban-traffic management system Trafﬁc-strategy Agperformance and minimizes the sys- e developers te c nt ra ﬁ es st Traf pe gi r fotem’s hardware requirements to ac- rm ancelerate the popularization of parallel ce Agent performancetraffic systems. Agent prototypeAgent-Based Traffic Typical trafﬁcManagement Systems Trafﬁc-strategy database Trafﬁc-strategy scenes testbed agent databaseAgent technology was used in traffic- Agent storage andmanagement systems as early as generation module Agent testing and training module1992, while multiagent traffic- performance en f Ask agent ag n o to move ts Agent fﬁc atiomanagement systems were presented tra bin mlater.5 However, all these systems fo- Cocus on negotiation and collaboration Database ofbetween static agents for coordina- te of knowledge age Mod nt- i sta dis ﬁedtion and optimization. 6–8 In 2004, nd n t ma ribut k a tio tas orta tio n p ionmobile agent technology began to at- fﬁc ansp ibu Tra tr str -di aptract the attention of the transpor- ent m Agtation field. The characteristics ofmobile agents—autonomous, mobile, Trafﬁc dataand adaptive—make them suitableto handling the uncertainties and in-constant states in a dynamic envi-ronment.9 The mobile agent movesthrough the network to reach con- Agent deployment and support moduletrol devices and implements appropri- Actual trafﬁc management system Artiﬁcial transportation systemate strategies in either autonomous or Figure 2. Overview of the main functional elements in the organization layer ofpassive modes. In this way, traffic de- Adapts.vices only need to provide an operat-ing platform for mobile traffic agentsworking in dynamic environments, as the carrier of the control strategies management agent (MA), three data-without having to contain every traf- in the system. bases (control strategy, typical trafficfic strategies. This approach saves In the follow-up articles, both the scenes, and traffic strategy agent), andstorage and computing capacity in architecture and the function of mo- an artificial transportation system. Asphysical control devices, which helps bile traffic control agents were defined one traffic strategy has been proposed,reduce their update and replacement clearly. The static agents in each layer the strategy code is saved in the trafficrates. Moreover, when faced with the were also depicted in detail. What’s strategy database. Then, according todifferent requirements of dynamic more, a new traffic signal controller the agent’s prototype, the traffic strat-traffic scenes, a multiagent system was designed to provide the runtime egy will be encapsulated into a traffictaking advantage of mobile agents environment for mobile agent. strategy agent that is saved in the traf-will perform better than any static Currently, Adapts is part of PtMS, fic strategy agent database. Also, theagent system. which can take advantage of mobile traffic strategy agent will be tested by In 2005, the Agent-Based Dis- traffic strategy agents to manage a the typical traffic scenes to review itstributed and Adaptive Platforms for road map. The organization layer, performance. Typical traffic scenes,Transportation Systems (Adapts) was which is the core of our system, has which are stored in a typical intersec-proposed as an hierarchical urban- four functions: agent-oriented task tions database, can determine the per-traffic-management system.10 The decomposition, agent scheduling, formance of various agents. With thethree layers in Adapts are organiza- encapsulating traffic strategy, and support of the three databases, thetion, coordination, and execution, re- agent management (see Figure 2). MA embodies the organization layer’sspectively. Mobile agents play a role The organization layer consists of a intelligence.January/February 2011 www.computer.org/intelligent 75
1,200 1,000 The function of the requires traffic control, Evaluation time (s) 800 agents’ scheduling and detection, guidance, mon- agent-oriented task de- 600 itoring, and emergency composition is based on subsystems. To handle the MA’s knowledge 400 the different states in a base, which consists of 200 traffic environment, an the performances of dif- urban-traffic management ferent agents in various 0 system must provide ap- 2 4 6 8 10 12 14 16 18 20 traffic scenes. If the ur- propriate traffic strategy Number of intersections ban management system agents. And to handle cannot deal with a trans- performance improve- Figure 3. The time required to run ATS and Adapts experiments portation scene with its on one PC. ments and the addition existing agents, it will of new subsystems, new send a traffic task to the traffic strategies must be organization layer for help. The traf- communicate with ATS to get traffic- introduced continually. So future fic task contains the information detection data and send back lamp- urban-traffic management systems about the state of urban transporta- control data. Both running load must generate, store, manage, test, tion, so a traffic task can be decom- and communication volumes increase optimize, and effectively use a large posed into a combination of several with the number of intersections. number of mobile agents. Moreover, typical traffic scenes. With knowl- If the time to complete the experi- they need a decision-support system edge about the most appropriate traf- mental evaluation exceeds a certain to communicate with traffic manag- fic strategy agent to deal with any threshold, the experimental results ers. A comprehensive, powerful deci- typical traffic scene, when the or- become meaningless and useless. As sion-support system with a friendly ganization layer receives the traffic a result, the carry capacity for ex- human-computer interface is an in- task, the MA will return a combina- perimental evaluation of one PC is evitable trend in the development of tion of agents and a map about the limited. urban-traffic management systems. distribution of agents to solve it. This In our test, we used a 2.66-GHz Thus, future systems must have the way, this system takes advantage PC with a 1-Gbyte memory to run following capabilities. of the strategy agent to manage a both ATS and Adapts. The experi- road map. ment took 3,600 seconds in real Computing Power Lastly, we set up an ATS to test per- time. The number of intersections The more typical traffic scenes used to formance of the urban-traffic man- we tested increased from two to 20, test a traffic-strategy agent, the more agement system based on the map and Figure 3 shows the time cost of detailed the learning about the advan- showing the distribution of agents. each experiment. When the num- tages and disadvantages of different ATS is modeled from the bottom up, ber of traffic-control agents is 20, traffic strategy agents will be. In this and it mirrors the real urban trans- the experiment takes 1,130 seconds. case, the initial agent-distribution portation environment.11 Because the If we set the time threshold to 600 map will be more accurate. To achieve speed of the computational experi- seconds, the maximum number of this superior performance, however, ments is faster than the real world, if intersections in one experiment is testing a large amount of typical traf- the performance is unsatisfactory, the only 12. This is insufficient to handle fic scenes requires enormous comput- agent-distribution map in both sys- model major urban areas such as Bei- ing resources. tems will be modified. jing, where the central area within Researchers have developed many the Second Ring Road intersection traffic strategies based on AI. Some New Challenges contains up to 119 intersections. We of them such as neural networks con- During the runtime of Adapts, we would need several PCs or a high- sume a lot of computing resources for need to send the agent-distribution performance server to handle the ex- training in order to achieve satisfac- map and the relevant agents to ATS perimental scale of several hundreds tory performance. However, if a traf- for experimental evaluation, so we of intersections. fic strategy trains on actuator, the ac- tested the cost of this operation. In Furthermore, a complete urban- tuator’s limited computing power and our test, traffic-control agents must traffic management system also inconstant traffic scene will damage76 www.computer.org/intelligent IEEE INTELLIGENT SYSTEMS
the performance of the traffic AI Intelligent traffic cloudsagent. As a result, the whole system’sperformance will deteriorate. If thetraffic AI agent is trained before mov-ing it to the actuator, however, it canbetter serve the traffic managementsystem. Rational traffic decisions and dis-tributions of agents need the support Trafﬁc participatorsof ATS, which primarily use agent- Trafﬁc managersoriented programming technology.Agents themselves can be humans,vehicles, and so on. To ensure ATSmirrors real urban transportation,we need large computing resources to Trafﬁc-strategy developers Urban-trafﬁcrun many agents. management systems Figure 4. Overview of urban-traffic management systems based on cloudStorage computing.Vast amounts of traffic data such asthe configuration of traffic scenes,regulations, and information of dif- urban-traffic systems at the time. clouds. The clouds’ customers suchferent types of agents in ATS need This way we can make full use of as the urban-traffic management sys-vast amounts of storage. Similarly, existing cheap servers and minimize tems and traffic participants existnumerous traffic strategy agents and the upfront investment of an entire outside the cloud.relative information such as control system. Figure 4 gives an overview of urban-performances about agents under dif- traffic management systems basedferent traffic scenes also consume a Intelligent Traffic Clouds on cloud computing. The intelligentlot of storage resources. Finally, the We propose urban-traffic manage- traffic clouds could provide traffic-decision-support system requires vast ment systems using intelligent traffic strategy agents and agent-distributionamounts of data about the state of clouds to overcome the issues we’ve maps to the traffic management sys-urban transportation. described so far. With the support tems, traffic-strategy performance to Two solutions can help fulfill these of cloud computing technologies, the traffic-strategy developer, and therequirements: it will go far beyond other multi gent a state of urban traffic transportation traffic management systems, ad- and the effect of traffic decisions to• quip all centers of urban-traffic E dressing issues such as infinite sys- the traffic managers. It could also management systems with a super- tem scalability, an appropriate agent deal with different customers’ re- computer. management scheme, reducing the quests for services such as storage• Use cloud computing technologies— upfront investment and risk for us- service for traffic data and strategies, such as Google’s Map-Reduce, ers, and minimizing the total cost of mobile traffic-strategy agents, and IBM’s Blue Cloud, and Ama- ownership. so on. zon’s EC2—to construct intelli- With the development of intelli- gent traffic clouds to serve urban Prototype gent traffic clouds, numerous traffic transportation. Urban-traffic management systems management systems could connect based on cloud computing have two and share the clouds’ infinite capabil- The former both wastes social re- roles: service provider and customer. ity, thus saving resources. Moreover,sources and risks insufficient capac- All the service providers such as the new traffic strategies can be trans-ity in the future. On the contrary, the test bed of typical traffic scenes, ATS, formed into mobile agents so suchlatter takes advantage of the infinite traffic strategy data ase, and traffic b systems can continuously improvescalability of cloud computing to dy- strategy agent database are all veiled with the development of transporta-namically satisfy the needs of several in the systems’ core: intelligent traffic tion science.January/February 2011 www.computer.org/intelligent 77
Application layer Standard application Standard interface programming interface Agent-generation service of urban-traffic management sys- Agent-oriented task Trafﬁc decision decomposition service testing service tems, support the running of agents Agent-management service and ATS test beds, and efficiently store traffic strategy agents and their Agent-testing service performances. Services Acknowledgments Platform layer We would like express our apprecia- Artiﬁcial trafﬁc systems tion to Fei-Yue Wang for his guidance and Services encouragement. This work was supported 3D editor Population generator Weather simulator Path planner in part by NSFC 70890084, 60921061, 90924302 and 90920305. Services VM VM VM VM Uniﬁed source layer References Virtual machine monitor 1. D.C. Gazis, “Traffic Control: From Hand Uniform data access Uniform computing access Collaboration and remote instruments Security Signals to Computers,” Proc. IEEE, vol. 59, no. 7, 1971, pp. 1090–1099. Management Services 2. F.-Y. Wang, “Parallel System Methods for Management and Control of Com- Fabric layer plex Systems,” Control and Decision, Agent vol. 19, no. 5, 2004, pp. 485–489. Algorithm performance library and parameters 3. F.-Y. Wang, “Toward a Revolution Trafﬁc scenes Storage resources Trafﬁc data Network resources Computing resources in Transportation Operations: AI for Complex Systems,” IEEE Intelligent Figure 5. Intelligent traffic clouds structure has application, platform, unified Systems, vol. 23, no. 6, 2008, source, and fabric layers. pp. 8–13. 4. F.-Y. Wang, “Parallel Control and Man- Architecture provide services to upper traffic ap- agement for Intelligent Transportation According to the basic structure of plications and agent development. Systems: Concepts, Architectures, and cloud computing, 12 an intelligent The unified source layer governs Applications,” IEEE Trans. Intelligent traffic clouds have four architec- the raw hardware level resource in Transportation Systems, vol. 11, no. 3, ture layers: application, platform, the fabric layer to provide infra- 2010, pp. 1–9. unified source, and fabric. Figure structure as a service. It uses virtu- 5. B. Chen and H. H. Cheng, “A Review 5 shows the relationship between alization technologies such as virtual of the Applications of Agent Technology the layers and the function of each machines to hide the physical char- in Traffic and Transportation Systems,” layer. acteristics of resources from users to IEEE Trans. Intelligent Transporta- The application layer contains all ensure the safety of data and equip- tion Systems, vol. 11, no. 2, 2010, applications that run in the clouds. ment. It also provides a unified access pp. 485–497. It supports applications such as agent interface for the upper and reason- 6. M.C.Choy, D.Srinivasan and R.L.Cheu, generation, agent management, agent able distribute computing resources. “Cooperative, Hybrid Agent Archi- testing, agent optimization, agent- All those will help solve information tecture for Real-Time Traffic Signal oriented task decomposition, and silo problems in urban traffic and Control,” IEEE Trans. Systems, Man traffic decision support. The clouds help fully mine useful information in and Cybernetics, Part A: Systems provide all the services to customers the traffic data. and Humans, vol. 33, no. 5, 2003, through a standard interface. Lastly, the fabric layer contains pp. 597–607. The platform layer is made of ATS, the raw hardware level resources 7. M.C.Choy, D.Srinivasan, and provided platform as a service. This such as computing, storage, and net- R.L.Cheu, “Neural Networks for Con- layer contains a population synthe- work resources. The intelligent traf- tinuous Online Learning and Control,” sizer, weather simulator, path plan- fic clouds use these distributed re- IEEE Trans. Neural Networks, vo1. 7, ner, 3D game engine, and so on to sources to cater the peak demand no. 6, 2006, pp. 1511–1531.78 www.computer.org/intelligent IEEE INTELLIGENT SYSTEMS
8. B.P. Gokulan and D. Srinivasan, “Distributed Geometric Fuzzy Multi agent Urban Traffic Signal Control,” IEEE Trans. Intelligent Transporta- tion Systems, vol. 11, no. 3, 2010, pp. 714–727. 9. N. Suri, K.M. Ford, and A.J. Cafias, “An Architecture for Smart Internet Agents,” Proc. 11th Int’l FLAIRS Conf., AAAI Press, 1998, pp. 116–120. 0. F.-Y. Wang, “Agent-Based Control1 for Networked Traffic Manage- ment Systems,” IE E E Intelligent Systems, vol. 20, no. 5, 2005, pp. 92 –96. 1. F.-Y. Wang and S. Tang, “Artificial1 Societies for Integrated and Sustainable Development of Metropolitan Systems,” IEEE Intelligent Systems, vol. 19, no. 4, 2004, pp. 82–87.1 2. I. Foster et al., “Cloud Computing and Grid Computing 360-Degree Compared,” Proc. Grid Computing Environments Workshop, IEEE Press, 2008, pp. 1–10.ZhenJiang Li is an assistant professor atthe Key Laboratory of Complex Systemsand Intelligence Science, Chinese Academyof Sciences. Contact him at email@example.com.Cheng Chen is a PhD student at the KeyLaboratory of Complex Systems and In-telligence Science, Chinese Academy ofSciences. Contact him at firstname.lastname@example.org.Kai Wang is a PhD student at the College ofMechatronics Engineering and Automation,the National University of Defense Technol-ogy (NUDT), Changsha, Hunan, China.Contact him at email@example.com. Selected CS articles and columns are also available for free athttp://ComputingNow.computer.org.