An Agent-based Model of Virtual
Experiment
Dazheng Wang
Ruimin Shen
Liping Shen
Department of Computer Science ,Shanghai Jiaotong Univ.
[Abstract]This paper analyzes the architecture of virtual experiment and discusses the
possibility of building a standard interface for it.We also introduces an agent-based VE
system ,which can integrates easily apparatus developed independently and using which
a teacher can just focusing his own field even without any programming technique.This
model may greatly eliminate many problems involved in software engineer and even
make distributed experiment possible.
[Index terms]Virtual Experiment,COM ,Agent
INTRODUCTION
With the development of the technology of computer and internet,various means to aid
people to study on computer are devised. Virtual Experiment(VE) belongs to such domain.VE is
an experiment simulated on computers for various reasons:to save money due to expensive
apparatus ,to ensure safety because of dangerous chemical powders and so on.However,an
interesting phenomena is that many VEs are developed in different entities though their functions
are same.
The components of VE,virtual apparatus,are difficult to reuse.On the other hand,the border
between teacher and developer is not clear and it is usually not possible for a teacher to design an
experiment with his own knowledge.This phenomena tells us a fact that we are wasting
unnecessary time and energy.So it is an urgent requirement for us to devise a standard interface of
virtual apparatus and a intelligent mechanism for teacher to design a VE without much
unnecessary effort.
Architecture of Virtual Experiment
There is no doubt that VE is a complex system,so different people may have different
perspectives on it.These people relevant to VE can be roughly divided into three
groups:teachers ,programmers and students. On the basis of top-down analysis,VE consists of
three parts:programming by programmers,designing by teachers and experimenting by students
(see Figure1).Though software design is crucial in VE and the teacher also take part in making
it,we should keep in mind the fact that not all of the teachers are programmers at the same
time,which implies that different people with different background shoud have different
responsibilities.A successful VE system must ensure designing an experiment as simple as to such
an extent that the teacher can design VE using no other expertise but his own instructional field .
Virtual
Experiment
Teacher
Designing
Experimenting
Student
Programming
Programmer
Figure 1 – Different Perspectives
Component Object Model
VE is made up of many virtual apparatus . To make them reusable and easy to integrate,the
best technology to adopt might be Component Object Model.The breakdown of a project into its
logical components is the essence of object-oriented analysis and design. That's also the basis for
component software, which is composed of reusable pieces of software in binary form (as opposed
to source code) that can be plugged into components from other vendors with relatively little
effort. It is important to realize that a component-based approach to software development does
not dictate the structure of an application. Rather, it is a model that makes possible the
programming, use, and independent evolution of binary software components. Components are
independent of the applications that use them as well as the programming languages used to create
them. Though COM has so many merits of independece,scripting skill is still needed so as to
design VE by gluing these virtual apparatus together. In order to eliminate the barrier between the
teacher and VE,we must devise a mechanism for components to adapt to and communicate with
the runtime environment of VE.The teacher can design VE by simple actions (such as drag and
drop)without bothering the script technique.
Virtual Apparatus
The aim of our VE framework is to gain the advantage of two features:
1. A standard programming interface
2. Minimum interference into the inner structure of each virtual apparatus
If we achieve this goal,we will never be trapped into the embarrassment of software
engineer,that is,modules are always difficult to integrate,there is usually many bugs in the
application,etc.The basic appearance of each virtual apparatus is shown in figure 2.
Virtual Apparatus
IDropSource
IUnknown
IAgent
Figure 2 – Virtual Apparatus
Figure 2 – Virtual Apparatus
Each virtual apparatus is implemented as a COM object with some interfaces exposed to the
ouside and a self-described XML file for registration.One of the interfaces ,called
IDropSource,must be inherited to enable the teacher to interact with the apparatus.The IAgent
interface will be discussed shortly. The special XML file is in fact a media for the virtual apparatus
and VE to interact and has two sections:one is a readable section for VE to obtain the information
of the COM object including its dynamic library file name,CLSID and IID,etc;the other is a
writeable section filled by VE ,where the object can get the necessary run-time information such
as its agent identity number , server name and port number.
Agent Structure
Since only minimum interference is done to each apparatus,it must do lots of work its
own ,such as changing its behavior according to the environment condition.This character makes
an apparatus look like an agent,so every one should implement the IAgent interface.Though
several characteristics of agents have been discussed in the literature, we assume that agents
exhibit the following three important properties in our modeling approach.
1.
Autonomy - an agent can make decisions about what to do based on its own state, without
the direct intervention of other entities.
2. Adaptation - an agent can perceive its environment,and respond to the changes in the
environment in a timely fashion.
3. Cooperation - an agent can interact with other agents through a particular
agent-communication language,and typically has the ability to engage in collaborative
activities to achieve its goal.
The agent wrapper of each apparatus is shown in figure 3.
Once
activated
by
the
environment,the agent goes into the
Agent
Agent
Deactivated
Activated
adaptation state.It sense the changing
condition of the outside by receiving
Adaptation
message from other agents.Then it judges
ReceivePMessage
the condition variables and sees if they are
ArrangeActions
relevant,otherwise it just discards
them.After the sensing phase,it can
produces some plan by arranging a series
ChangeAppearance
of actions,which may result the changed
appearance or its inner states.The agent
SendPMessage
can also give feedback to the environment
by changing the experiment parameters.
Figure 3 – Agent Structure
The basic element of sensing
and
feedback states is a message queue,with which the agent can responds asynchronously without
blocking other agents and the environment.The agent comes to the end of its life cycle if
deactivated by the outside.
Communication
Only a minimum number of functions in an agent are exposed to the environment,such as
Activate and Deactivate,but they cannot be accessed by other agents.The messages between agents
are passive ,i.e.,they do not contain methods because agents never know each other’s
structure.Once activated ,an agent just sends out massages containing measurement parameters
according to its own state transitions,which are sent to another agent connected to it or to the
environment.The parameters might be temperature,pressure and so on.Because the message is
different from traditional method invocations,we name it as passive message or PMessage. A
sample PMessage may looks like this(See Figure4). The communication mechanism is
implemented on the basis of socket technique. At the beginning of its life cycle ,every agent must
create a listening socket with the server address and port number assigned by the environment on
registration. Those who want to communicate with it must bind to that address and port
number.Such a email-like communication mechanism can make distributed experiment possible. If
agents want to have a conversation ,they must send their identity numbers to each other via
run-time environment.Figure 5 shows such scenario.
Run-time
Environment
From:30001
To: 30002
Category:physics
Name:temperature
Unit:FD
Value:45
4: ID1
2: ID1
3: ID2
1: ID2
6: PMessage
Agent1 :
coclass
Figure 4 – Sample PMessage
Agent2 :
coclass
5: PMessage
Figure 5 – Conversation scenario
Conclusion and Futrue Work
Since our virtual apparatus component is highly autonomous and possesses a clean programming
interface,potential trouble for integrating them into a virtual experiment system is avoided.At the same
time,we have achieved the feature for teacher to design an experiment with little effort.Futrue work
will focus on the concurrency and responsiveness of
attention to the analysis of heterogeneous
our agent model.We will also pay much
experiment platforms and realize the distributed virtual
experiment system.
References
[1] [Burmeister, 1993] B. Burmeister, A. Haddadi, and K.Sundermeyer. Generic, Configurable, Cooperation
Protocols for Multi-Agent Systems. Proceedings MAAMAW'93, pp. 157-171, Springer, 1993.
[2]
COM Specification http://www.microsoft.com
[3] [Brazier, 1998] F. M. T. Brazier, C. M. Jonker, and J.[Odell, 1999] J. Odell. Agent Technology, green
paper,produced by the OMG Agent Working Group, ed., 1999.
[4]
[Garijo, 1999] F. J. Garijo, and M. Boman. Multi-Agent System Engineering. Proceedings of
MAAMAW'99.Springer, ed., 1999.
[5] [Jonker, 1997] C. M. Jonker, and J. Treur. Compositional Verification of Multi-Agent Systems: a Formal
Analysis of Pro-activeness and Reactiveness. Proceedings of International Workshop on Compositionality
(COMPOS'97), Springer, 1997.
[6] [Bauer, 1999] B. Bauer. Extending UML for the Specification of Interaction Protocols. submission for the 6th
Call for Proposal of FIPA and revised version part of FIPA 99, 1999.