Get documents about "Towards Building an Intelligent Traffic Simulation
Document Sample
Towards Building an Intelligent Traffic Simulation Platform
Jian Caoa Minglu Lia Linpeng Huanga Ren Qinshenga Ying Lia, b
a
Department of Computer Science, Shanghai Jiaotong University, Shanghai, 200030, P.R.China
b
Computer Science and Technology School, Soochow University, Suzhou, China.
a
{cao-jian, li-ml, huang-lp, ren-qs, jessieLee}@cs.sjtu.edu.cn
Abstract providing decision support by applying traffic
simulation is also listed as one of the research topics.
Traffic congestion has become a major concern for Traffic simulation has already gained much
many cities throughout the world. Simulations provide attention since it can help government optimize their
useful tools for engineer to plan traffic systems and infrastructure investments and traffic management
government to make decisions. The microscopic traffic policies in a cost-effective manner. Therefore, in the
simulation approach defines the behavior of each last a few years, and several packages such as
interactive object in the traffic network, such as a CORSIM, CONTRAM, CORFLO, PARAMICS, have
vehicle and a traffic light through an individual model been implemented [1].
so that we can observe the detail traffic information of There are two different approaches used to simulate
a scenario. Since microscopic model requires large the traffic in a network. The first one, the macroscopic
computational power and data storage power, new approach is based on an analogy between traffic flow
simulation system architecture is needed. In the paper, and a real fluid flow. The second one, the microscopic
the background of this research work is introduced. approach simulation deals with the individual
The simulation platform which combines grid, agent behaviors of each vehicle [2]. The macroscopic
and HLA is given. Some considerations about approach presents a higher level of abstraction than the
microscopic traffic simulation models and technical microscopic one so that it can model large scale traffic
issues of this platform are also discussed. network and needs less data and computational power.
However, macroscopic simulator simplifies the model
1. Introduction too much and many complex factors and details can
not be modeled and analyzed. On the other hand, the
Traffic is a major concern in modern cities microscopic approach defines the behavior of each
throughout the world, especially in large city like interactive object in the traffic network, such as a
Shanghai. Although Shanghai government invests vehicle and a traffic light through an individual model
heavily into building roads, improving their traffic so that we can observe the detail traffic information of
control systems in recent years, the challenge still a scenario. Although the microscopic approach can
becomes larger with more and more people buy their provide more information, it needs large data and more
private cars. computational power and to simulate a relative large
From year 2004, under the support of a long-term area the computational time can become prohibitive.
project, Shanghai Grid, which is sponsored by the Since the processing capability of computer
Science and Technology Commission of Shanghai increases rapidly, the use of traffic micro-simulation
Municipality, a intelligent traffic information service models in traffic operations, transportation design, and
grid are developed to integrate traffic information, transportation planning has become widespread. But
share traffic data and traffic resources so that to to develop a more useful traffic micro-simulation
provide better traffic services to traffic participators, platform is always a challenge since we can add more
help remove traffic bottlenecks and resolve traffic details into the model if we have stronger computing
problems. The first stage of this intelligent traffic power. Some new technology such as grid offers great
information service grid project was finished in 2005 opportunities to explore more complex models in
and the second stage began from 2006. With the aims traffic simulation.
of developing more useful traffic services and This paper introduces an ongoing research, which
improving the usability and reliability of this grid, combines grid, agent and HLA to support microscopic
traffic simulation.
The paper is organized as follows. Section 2 gives a Shanghai has become the largest economic center
belief introduction to the background of this research and an important port city in China, with a land area
work. Section 3 discusses related supporting covering 6340 km2 and a population of 16 million
technologies. In Section 4, a simulation platform is people. With the rapid development of economic and
introduced. Section 5 presents some considerations the increasing number of automobiles, the problem of
about microscopic traffic simulation models and urban traffic congestion has become more and more
technical issues. Section 6 discusses related works and serious. To solve such problems, Shanghai government
finally Section 7 concludes the research and points out puts its focus not only on road infrastructure
some future works. construction, but also on transportation control systems.
These systems play important roles in solving the
2. Research Background traffic problems in Shanghai, acting as subsystems of
ITS.
2.1 Shanghai Grid In using these systems, one problem has emerged
gradually. The design of these systems does not
consider the interoperation among each other. These
In 2003, under the support of Science and
systems belong to different government agencies, use
Technology Commission of Shanghai Municipality, a
different technologies and the traffic data cannot be
long-term project called Shanghai Grid was started.
shared among them. Another problem is how to store,
Shanghai Grid aims to construct a metropolitan-area fuse, and utilize the transportation information data
Information Service Grid (ISG) and establish an open (TID) in these systems. TID is fundamental to ITS, as a
standard for widespread upper-layer applications from big city, the amount of TID of Shanghai in each system
both communities and the government. It is one of five are huge. Even more, different systems use different
top grand Grid projects in China. The first test bed is ways to store their data. It is difficult to provide a good
based on the current four major computational way to interoperate among these systems.
aggregates and networks in Shanghai, including Shanghai government has already noticed the
Shanghai Supercomputing Center (SSC), and various weakness of the non-interoperation of these systems. In
campus supercomputer centers in Shanghai Jiao Tong order to provide better services to citizen, further
University (SJTU), Tongji University (TJU) and reduce the traffic congestion, provide real-time traffic
Shanghai University (SHU). It is planned to enable the information to decision makers, it launches the ITIS
integration of heterogeneous and distributed resources project, which aims to build a platform to integrate
seamlessly and transparently. Shanghai Grid has various transportation systems as a whole. ITIS will be
connected several major grid nodes to form a 0.6 based on Shanghai Grid.
Tflops aggregate computing power, a 4 TB aggregate ITIS comprises five sub-projects, including
storage power and a sophisticated information service 1)Development of open protocols and standards for
environment [3]. transportation information service applications;
The aim of shanghai government is to use the grid 2)Research on Grid supporting platform;
technology to construct a basic infrastructure for e- 3)Transportation resources integration; 4)Dynamic
science, e-business, e-education, e-government and e- parallel traffic simulation; 5)implementation of
life, as the basic facilities of the city, similar to intelligent traffic information services.
transportation and communication systems, water and As one sub-project of ITIS, the aim of traffic
power lines. Therefore Shanghai Grid as an simulation is to construct a platform to support
infrastructure will fully use the existing techniques and research on microscopic traffic phenomenon and
resources to provide rich functionality of information explore how to define appropriate traffic control rules
services. Currently, several applications have been under different situations, especially under the
developed and put into use in the Shanghai Grid, such emergent situations in a dedicated area. This kind of
as computational fluid dynamics, medicine image simulation needs large amount of computational power
processing, drug discovery Grid, et al. The core of and also the storage power, which can not be satisfied
Shanghai Grid is the SHGOS, which is a middleware be a single computer. Therefore, we will base this
providing services and tools to satisfy the needs of research on the technologies such as grid, HLA and
building the ISG. Moreover, the SHGOS hides the distributed agent systems which can be used to
complexity of the Grid techniques for developers construct a scalable system.
building grid applications.
3. Related Technologies for Traffic
2.2 Intelligent Traffic Information Service Grid Simulation
3.1 Agent and Traffic Simulation not reflect complex behaviors of entities and there are
no direct interactions among grid services. From the
The concept of agent is now broadly used not only viewpoint of simulation, the traffic model can not map
as a model for computer programming units displaying to the grid directly.
certain kinds of characteristics but also in a more
abstract and general way, as a new metaphor for the 3.3 HLA and Traffic Simulation
analysis, specification, and implementation of complex
software system [4]. In order to support development The HLA (High Level Architecture) [8] is an
of agent based systems, in recent years, a number of industry (IEEE-1516) standard for modeling and
agent infrastructures have been developed to facilitate simulation. It is increasingly being used in various
the implementation of multi-agent systems. Some simulation areas, including education, training,
systems such as JADE [5] provide the bare bones of analysis, engineering, entertainment and games. In the
distributed communication. HLA, a distributed simulation is called a federation,
There is by now some agreement that multi-agent and each individual simulator is referred as a federate,
simulation may be a viable technology for microscopic one point of attachment to the Runtime Infrastructure
traffic simulation. In a multi-agent based traffic (RTI). A federate can be a computer simulation; it can
simulation system, each traveler, and potentially each also be a physical device, a passive data viewer or an
entity of the simulation, such as traffic lights or interface to a human participant.
variable message signs, is represented as individual While the High Level Architecture (HLA) enables
objects or “agents,” which make independent decisions interoperability and the construction of large-scale
about their actions. distributed simulations using existing and possibly
Although the concept of applying agent into traffic distributed simulation components, it does not provide
simulation is quite straightforward, there are some support for simulation application development, nor
difficulty implementation issues. The first issue is how does it provide any mechanism for managing the
to maintain the causality, which is the basic feature of resources where the simulation is being executed.
parallel simulation. The second one is the Obviously, agent, grid and HLA can compensate
computational power requirements are very large and each other in some degree when they are applied to
the third one is the communication overheads may be microscopic traffic simulation. This project will
too heavy if each agent should exchange information explore how to combine these technologies into a
with each other. unified framework.
3.2 Grid and Traffic Simulation 4. The Traffic Simulation Platform
Architecture
In order to solve complex scientific problems,
geographically distributed and heterogeneous resources Fig.1 shows an intelligent simulation platform
should be connected together through the high-speed which based on agent, RTI and grid. The user can
network. Grid, which provides virtual organization design the simulation model through the tool of
with capabilities to solve the problems cooperatively, modeler. The generator will produce and remove
is put forward to meets these more and more critical agents by using some services provided by the agent
requirements [6]. As a new infrastructure, grid is platform according to the simulation model
extending its role as a high performance computing dynamically when simulation begins. These agents
environment and bringing deep impacts on the human represent the entities of the traffic, such as passenger,
life and the society [7]. vehicle, police or road. When an entity agent is created,
Grid technologies also provide exciting new it will be given the behavior model according to the
opportunities for large scale distributed simulation, simulation model but with different characters which
enabling collaboration and the use of distributed reflects the individual differences. All agents are
computing resources, while also facilitating access to deployed into distributed computers which are
geographically distributed data sets. Since traffic connected through an agent platform. There are a set of
simulation needs resources, grid can be applied to this special agents called coordination agents, which play
application. the role of federates connecting to the RTI. Each entity
Since grid is a platform for resource sharing among agent should belong to a coordination agent and it only
dynamic virtual organization, some necessary communications with the corresponding coordination
simulation facilities are not provided in grid and should agent. The entity agent reports its current status to the
be realized on the top of grid services. Comparing with corresponding coordination agent and it also obtains
agent technology, the grid services are passive, it can
the environment information from the coordination destination can be defined as the objective of a
agent. Coordination agents exchange information passenger agent.
through the RTI and the simulation times of
coordination agents are advanced through time 5. Simulation Model and Simulation
management service of RTI. Mechanism
The computational tasks of entity agent can be
allocated to grid services and the grid service can also 5.1 Entity Agent Model
collect real data from the real traffic system to store
into the model base. Within the traffic environment, the entity which has
A “perception-interpretation-action” model is different behaviors under different situations can be
adopted in that an agent continuously assesses or modeled as a separate agent. These agents include
“senses” the surrounding environment from the passenger agent, bicycle agent, vehicle agent, road
coordination agent and makes decisions based on its agent, traffic light agent, police agent and control
decision model in a proactive fashion. Each agent center agent.
includes a sensor so that an agent can analyze the 1) Passenger Agent: the behaviors of passengers have
environment. Agent’s actions are represented in terms a great impact on the traffic flow especially when some
of decision rules. When a situation is perceived, an passengers do not obey the traffic rules and this is
agent activates a decision rule to produce an action. often the case in China. Human individuals are
The choice of a decision rule is determined by the different from each other by age, body dimension,
situational cues and the agent’s status (i.e., perceived motility, and personality. Therefore the generator will
importance, uncertainty and urgency) at that moment. create the individual passenger randomly according to
For a mobile entity, such as passenger, bike or vehicle, the models defined. The sensor of passenger agent can
its corresponding agent’s behaviors can be categorized obtain the geometrical distance from the intersecting
into several hierarchical layers (from simple to object, and also determine the type of object. The
complex), for example, the locomotion, steering, social actions of passenger agent includes walking forward,
and objective layers. The behaviors on a higher layer running forward, stopping, side-shifting, turning, and
are constructed using the behaviors from a lower layer.
Grid
Agent Agent Agent Agent Agent Agent Modeler
Generator
Visualizer
Coordination Agent Coordination
Agent Platform Agent
Federate
Federate Federate
……
RTI
Fig.1 An Intelligent Simulation Platform
The locomotion layer defines the basic actions the moving backward. The steering behaviors include
agent can take. The steering layer defines a group of random walk, collision avoidance, seek, negotiation,
actions or a sequence of actions that the agent can take. target following, Social behavior includes competitive,
The social layer defines the group behavior this agent queuing, herding [9]. The objective behavior includes
follows. For example, the passenger or the vehicle passing this area, going to some places in this area. The
follows other passengers or vehicles when they meet decision rules include avoiding collision, walking
the emergency. The objective layer defines what kind faster as possible, following traffic light, et. al..
of aim this agent has when it enters into this area. For 2) Bicycle Agent: Since a large part of people use bike
example, walking from the current location to the as their transportation tool in China, we should
consider the affections of bikes. The model of bike is actions based on decision rules and reports the renewed
some kind of mixture of passenger and vehicle agent status to the coordination agent. When received the
models. Some road has a special lane for bike while responds from all entity agents, it notifies entity agents
some road has no this lane. The locomotion actions of it managed to obtain the information again.
bicycle agent include driving forward, speeding, How to design coordination agent and allocate entity
slowing down, stopping, side-shifting, turning. The agents to the coordination agent is a very important
steering behaviors include random walk, collision research topics. We can divide a road into several
avoidance, seek, negotiation, target following, Social pieces and each piece is allocated to a coordination
behavior includes competitive, queuing. The decision agent, other agents which locate in this piece of road
rules include voiding collision, walking faster as will connect to the corresponding agent. Then an
possible, following traffic light, et. al. important concern is some agents, such as passengers,
2) Vehicle Agent: The vehicle model is researched bikes, vehicles, moves from one place to another. Their
extensively and there are already quite a few move is controlled by themselves. Therefore, each
mathematic models to describe the behaviors of a coordination agent should have a global map and it
vehicle. What we have to do is to map these models to knows which coordination agent is responsible for
the structure of agent. The sensors of vehicle agent will which pieces of road so that it can transfer a mobile
obtain the speed of itself, the vehicles around itself and agent to another one. Another issue is the agents in the
traffic signals. The locomotion actions include driving border of road pieces affect each other. Since agents
forward, speeding up, slowing down and stopping. The staying in different road pieces can not exchange
steering behaviors include collision avoidance, seek, information directly, we should define a mechanism to
negotiation, target following. The social behaviors allow them exchange information through their
includes competing, queuing. The decision rules corresponding coordination agents.
include voiding collision, walking faster as possible,
following traffic light, et. al. 6. Relate works
3) Road Agent: this agent is relatively simple because
it has less intelligence and only act as an information Simulation based on multi-agent systems has
broadcasting place. The road agent may provide some gaining more and more attentions in recent years.
guide information to the vehicles and it also calculates Despite of many applications, some researches are
the status and reports it to the control center. focusing on how to develop such a system. For
4) Traffic Light Agent: traffic light agent can be simple example, the synchronization of the participating
or complex. For a passive traffic light, it only changes agents with respect to the global simulation time is a
the light color according to the predefined rules. But necessary requirement of testing and simulation of
the traffic light can be complex one, and it can change process flows within multi-agent systems. A design
the rule according to the commands coming from the proposal and a service implementation for testing and
control center. simulation is presented, which takes care of the special
5) Police Agent: Police agent can change the road rules requirements imposed by multi-agent settings [10].
according to the commands from the control center. It Time service is implemented as a FIPA-compliant
affects the behaviors of passengers, bikes and vehicles. agent, and can be used to couple heterogeneous
6) Control Center Agent: The control center will subsystems implemented on different agent platforms.
monitor the situation and send the commands to the Another approach to support agent based simulation is
police, traffic light and the road agents. The sensor of to integrate agent platform and HLA RTI together [11].
this agent should collect all the data from the road Although HLA provides some services that distributed
agents. Some complex decision making models can be and parallel simulations needs, there are still some
defined in this agent and the user can also change the important issues should be addressed, especially for
decisions of this agent interactively. large scale simulation system. For example, the
communication problem is discussed in [12]. In our
5.2 Simulation Mechanism proposed structure, agent platform and HLA RTI are
also combined together, those problems, such as
For a complex multi-agent based distributed and communication still remains. A more challenge
parallel simulation system, three major concerns have problem is the connecting relationships among agents
been pointed out in Section 3.1. Our solution is based are changing in simulation process and we are also
on coordination agent, which connects all related entity carrying out research on the mechanism.
agents together and also maintains the corrections of Traffic simulation based on agent has been
time sequences. Each entity agent obtains the data researched extensively. For example, there is an
from the coordination agent, and then triggers the
ongoing research to implement a multi-agent [5] Jade - Java Agent DEvelopment Framework,
simulation for all of Switzerland, which, with about 7 http://jade.tilab.com/ , 2006.3
million inhabitants, also serves as a proxy for a large
metropolitan area. In this system, each traveler is [6] Foster, I., Kesselman, C., Tuecke, S., “The Anatomy of
the Grid: Enabling Scalable Virtual Organization”,
modeled as an agent, which makes independent
International Journal of Supercomputer Applications,
decisions about their actions. In [13], a research which 2001.3, Vol. 15(3), pp200-222
is based on swarm intelligence to observe emergent
behavior patterns in traffic is introduced. Comparing [7] Foster, I., Kesselman, C., et.al., “The Anatomy of the
with these works, our system will use a layered Grid: Enabling Scalable Virtual Organizations”,
behavior model so that we wish we can find more International Journal of Supercomputer Applications, 2001,
interesting results when we define more complex Vol. 15(3), pp200-222,
models.
[8] Katherine L. Morse, Mike Lightner, Reed Little, Bob
Lutz, Roy Scrudder, “Enabling Simulation Interoperability”,
7. Conclusions and future work Computer, 2006.1, Vol. 39(1), pp. 115-117
Traffic control in a city like Shanghai is a major [9] Xiaoshan Pan, Charles S. Han, Kincho H. Law, “A Multi-
concern of government. The information technology is agent based Simulation Framework for the Study of Human
widely applied in transportation systems. The emergent and Social Behavior in Egress Analysis”,
new technologies provide new possibilities to explore http://eil.stanford.edu/egress/publications/
innovative applications for traffic control. The agent, ASCE_2005_Pan_Han_Law.pdf, 2006.3
grid and HLA technologies can be applied together to
[10] L. Braubach, A. Pokahr, W. Lamersdorf, “A Generic
construct a new generation of traffic simulation Simulation Service for Distributed Multi-Agent Systems”,
system. The agent model provides new ways to allow Cybernetics and Systems, 2004, Vol. 2, Vienna Austria, pp.
considering different complex behavior patterns. 576--581
We have defined the architecture of simulation
system, the next step is to develop such system and we [11] Fang Wang, Stephen John Turner, Lihua Wang, “Agent
wish until the end of year 2006 we could have one Communication in Distributed Simulations”, MABS 2004,
prototype system. pp11-24
[12] Bryan Raney, Kai Nagel, “An Agent-Based Simulation
Acknowledgements: Model of Swiss Travel: First Results”,
http://www.strc.ch/Paper/raney.pdf , 2006.3
This paper is supported by ShanghaiGrid grand
project of Science and Technology Commission of [13] Joanne Penner, Ricardo Hoar, Christian Jacob, “Swarm-
Shanghai Municipality (No. 05DZ15005) and Chinese Based Traffic Simulation With Evolutionary Traffic Light
NSF Project (No. 60503041). Adaptation”, http:// www.swarm-design.org/SwarmDesign/
Papers/ASM-2002-363-099.pdf, 2006.3
References
[1] Skabardonis, A., May, A. , “Simulation Models For
Freeway Corridors: State-Of-The Art And Research Needs”,
Transportation Research Board 77th Annual meeting,
January 1998, Washington D.C
[2] Laurent Magne, Sylvestre Rabut, Jean-François Gabard,
“Towards an Hybrid Macro-micro Traffic Flow Simulation
Model”, http://www.cert.fr/dcsd/cd/MEMBRES/magne/
slc2000.pdf, 2006.3
[3] Ying Li, Minglu Li, Jiao Cao, et.al., Towards Building
Intelligent Transportation Information Service System on
Grid. APWeb Workshops 2006: 632-642
[4] Oliveira Eugénio, Fischer Klaus, Stepankova Olga,
“Multi-agent Systems: Which Research for Which
Applications”, Robotics and Autonomous Systems, 1999.4,
Vol.27(1-2), pp. 91-106
