7.5 Examples

7.5.1 Agentcities

Agentcities ^[4] is a worldwide initiative designed to help realize the commercial and research potential of agent-based applications by constructing a worldwide, open network of platforms hosting diverse agent-based services. The ultimate aim is to enable the dynamic, intelligent and autonomous composition of services to achieve user and business goals, thereby creating compound services to address changing needs.

^[4] http://www.agentcities.org/

The initiative will build on a wealth of innovative technologies including agent technology, semantic Web technologies, UDDI discovery services, e-business standards and grid computing. Application areas already envisaged range from e-health and e-learning to manufacturing control, digital libraries, travel services, and entertainment services.

The Agentcities network is designed to act as a distributed testbed for experimenting with agent technology and composable services. It allows the creation of a common resource for developers wishing to collaborate with each other and link up their agent systems and services. Agentcities also provides a benchmark environment to validate and test compliance to relevant technology standards and provide input to the standards themselves . It can also act as a focus for discussion of next -generation information networks as well as the development of services, technologies, and methodologies (see Table 7.3).

While infrastructure (messaging, directories, etc.) is necessary to create the network, the objective of the Agentcities project is not simply to deploy infrastructure. Many of the real challenges of deploying and using such a network lie in how diverse services can discover each other, the development and usage of a semantic framework, and how coordination can be achieved between heterogeneous systems.

The primary objective of Agentcities is, therefore, to create a rich, open environment to explore these questions. Although a number of the projects underpinning Agentcities take the area of travel, tourism, and entertainment as their application domain, there is no restriction on the applications that could be deployed in the Agentcities network.

The main example of developing a platform in the network to model the services available in a town or city (hence, the name Agentcities) simply provides a convenient domain focus to begin tackling the problems of semantics, ontology, and dynamic service composition in manageable proportions (see also Figure 7.4). Other groups are already considering very different application domains and how to exploit Agentcities in their own way. Interest groups that are already forming include:

Figure 7.4. The Agentcities Web Site

• Travel, tourism, and entertainment services Focal application area of the EU Agentcities research project.

• Business services Marketplaces, payment systems, transactions and catalogue services.

• Coordination technologies Coordination media and shared coordination methods .

• Medical and healthcare services Distributed services for organ transplantation , access to patient medical records, and local emergency services linked with existing projects and new activities.

• Manufacturing and supply chain integration Using the Agentcities infrastructure as a substrate for coordinating distributed manufacturing processes and supply chain integration.

• Security services Using the Agentcities infrastructure as a testbed for analyzing and beginning to address the security needs of such open, heterogeneous environments.

• e-Learning Distributed agent-based tutoring systems.

• Wireless applications Seamless interaction between wireless and wire line agents to dynamically compose service based on user location.

• Personalization Dynamic composition of user services to suit individual tastes.

Table 7.3. Overview of Agentcities

Agentcities is based on these principles:

• Consensual Standards Communication and interaction in the network will be based on publicly available standards, such as those developed by the Foundation for Intelligent Physical Agents (FIPA) and the W3C.

• Open Source Although commercial technologies are not discouraged, Agentcities will promote freely accessible open source implementations to ensure free and open access to the network.

• Open Access Any organization or individual can set up their own Agentcity in the network to host their own agent services, provide access to them and access those deployed by others.

• Shared Resources People accessing agent-based services in the network, such as directory, naming, and application services, are encouraged to add their own services to extend the utility and diversity of the services available to the community.

The Agentcities project will have a significant impact on the global deployment of agents and will provide a useful resource for the development of the next generation of networked systems. However, Agentcities is not attempting to be the panacea of agent or network programming since much hard work is involved in developing semantic frameworks and content languages for agents to communicate. Therefore, the role of the Agentcities project is to stimulate this process and encourage participants to think in an open context and to envisage their systems in the context of a worldwide environment. The process of connecting an increasing number of diverse agent systems will teach us much about which details matter when it comes to creating true interoperability, not just at the syntactic layer, but also at the semantic layer.

7.5.2 Hive

Hive is one of the first distributed agents platform built by MIT ^[5] to provide the integration of ubiquitous computing with wearable devices. It is a decentralized system for building applications by networking local resources. The key abstraction in Hive is the software agent: Applications are built out of an ecology of multiple agents interacting over the network.

^[5] http://www.hivecell.net/

From a programmer's perspective, each Hive agent is a distributed Java object and an execution thread, which also implements the following properties:

• Autonomous Agents can be sent into a system and entrusted to carry out goals without direct human micromanagement.

• Proactive Because agents have their own threads, they can act independently of other running agents. They encapsulate computational activity.

• Self-describing An ontology of agent capabilities can be used to describe and discover available services. Hive agent descriptions consist of both a syntactic description (represented by the Java class of the agent) and a semantic description written in the Resource Description Format (RDF, see Chapter 6).

• Interactive Agents can work together to complete a task. Hive agents can communicate both through an asynchronous event/subscriber mode and through Java RMI (Remote Method Invocation). Agent communication is completely peer-to-peer, so an agent might both send and receive at different times.

• Mobile Agents can move from one physical device to another.

Along with agents, the Hive architecture defines so-called "shadows." Shadows are the low-level drivers for the physical resources of a particular object. For security, only local agents can access a particular shadow. All remote agents that wish to access local resources must go through local agents. Finally, Hive defines the concept of a "cell," which encapsulates a group of agents. The cell is essentially a Java process that implements an environment within which the agents operate . Generally, there will be one cell per local object or wearable computer, though this is not a hard and fast rule. A wearable computer is a computer that is subsumed into the personal space of the user, controlled by the user, and has both operational and interactional constancy. Agents are free to communicate and move between cells . Hive also provides a graphical interface to view and manipulate agent interactions. The interface is itself just one more Hive agent that receives and displays agent events.

Hive supports the discovery of new agents through two kinds of lookups. Agents can be queried both based on their syntactic description and on their semantic description. Semantic descriptions include information such as a resource's owner or the room in which it lives.

Using this infrastructure, several agents have been created to make resource finding simple for wearable computer users. MIT has created a "resource finder agent" that receives location events and produces sets of agents whose semantic descriptions match that location. These agent sets can then be winnowed further by other resource finder agents that are looking for specific services like stereo equipment or large-screen monitors . To bootstrap the initial list of cells, a resource finder agent contacts a known master cell listing agent. The creation of new agents is announced to subscribing agents in the form of events, or a special cell managing agent can be queried.

In this system, each wearable computer is simply another Hive cell in a decentralized Hive network. Sometimes, the wearable is the interface to an external service; sometimes, the wearable is a service provider for an agent in the environment (or on another wearable); and sometimes, the wearable and environment are interacting.

On Hive, several agents have already been implemented. The following provides a very short overview on their functionality. The first one is the automatic diary, which basically stores the location information about a user. Whenever the wearable user types in a note or idea, that note is automatically timestamped and tagged with the location where that note was made.

Besides the automatic diary, MIT has implemented a DJ agent, which plays the right music in a given room for a certain occasion. If a person enters the room and no music is running, the agent will look at the personal preferences of that user and select the appropriate music.

To make sure that the music does not annoy other people in the room, the DJ agent implements resource management policies to ensure fairness. In the default case, a DJ takes requests sequentially and plays one request for each agent that has a song. Thus, people's requests will be played in a round- robin fashion, one request per person.

In addition to this, a remembrance agent has been implemented. As an example, imagine a wearable that acts as a tour guide at a museum. As the visitor moves through various exhibits, extra information about each site is presented on her wearable. Because this information is location specific, it is more easily maintained if it resides in the museum databases and is sent to a visitor's wearable only when she enters the exhibit area.

Besides providing services to the user, the architecture can also make the wearable act as a service provider. The so-called "Where's Brad?" agent produces a map that shows the current location of a wearable user. This agent uses the same resource discovery tools as do the agents on the wearable, except now instead of finding agents associated with a location, they find agents associated with a particular person.

7.5.3 OnStar

OnStar, ^[6] which connects drivers to a 24- hour on-call center for emergency services or help finding a restaurant or hotel, is one of the new services GM is offering. It is one of the most popular in-vehicle communications system in the United States, with more than 2 million subscribers. It is a subsidiary of General Motors ^[7] and provides service for many car models, including GM, Lexus, Audi, Saab, and Volkswagen.

^[6] http://www.onstar.com/

^[7] http://www.gm.com/

OnStar can be considered as a collection of Web services and agents that integrate in an open marketplace on a portal. The portal knows about capable resources (agents) for a certain task and can bring these into play whenever required. This portal can be seen as a concierge that would be the natural mode for me-centric access to services in various environments. The system can, for example, alert an OnStar operator when an air bag deploys, but also notify operators about accidents that did not trigger an air bag deployment and send more information about the crash. Crash sensors in the front and rear bumpers and on both sides of the vehicle can tell where the vehicle was hit and the speed and force of impact. This technology is especially helpful for accidents in rural areas, where there may not be witnesses and the victims may not be immediately discovered .

OnStar operators can also locate a stolen vehicle, remotely unlock the doors, provide roadside assistance, perform remote diagnostics, give directions, and even make dinner reservations . Similar services have been set up by DaimlerChrysler ^[8] (TeleAid and Dynaps) and Fiat ^[9] (Connect).

^[8] http://www.daimlerchrysler.com/

^[9] http://www.fiat.com/

In the future, the information may also be transmitted electronically to 911 centers, first responders, and hospitals using secure Internet connections. The system may also one day be able to tell how many people were in the vehicle, whether they were using seat belts, and other information that helps emergency responders anticipate injuries before they arrive at the accident scene.