Applications

MedPAge (Medical Path Agents) is a research project, focusing on treatment scheduling for patients in hospitals. Scheduling and coordinating patients in hospitals is faced with a high amount of complexity due to the inherent dynamics of the processes and the distributed organisational structure of hospitals. Multi-agent technology facilitates solutions to these problems, as the autonomy of agents allows to maintain the integrity of the existing organisational structure of hospitals. Furthermore, agents are able to react flexible to changes and disturbances (e.g. emergencies and complications) through pro-activeness and reactiveness.

The MedPAge project is part of the DFG-funded German Priority Research Programme SPP-1083: "Intelligent Agents in Real-World Business Applications" which was initiated in 1999. All SPP projects are carried out as so-called tandem projects: One partner from business economics and one from computer science. Our project partner is the Business Informatics Chair at the University of Mannheim. In addition, we work together with computer science partners from other SPP projects, most notably with the RWTH Aachen and the University of Würzburg.

In MedPAge patients and hospital resources are represented as autonomous agents with individual goals. For coordination we conceived a market-based mechanism called MedPaCo (Medical Path Coordination), in which the patient agents negotiate with each other - based upon individual health state dependent cost functions - over the scarce hospital resources. A hospital simulation environment allows the benchmark of different coordination mechanisms including the current practice in hospitals. In contrast to the resource agents that only see the patients as entities to be treated, the patient agents merely see the medical actions as tasks that need to be performed. Due to these opposing forces, the patient agents ensure that the resource agents also consider the treatments of the patients outside their unit (without any explicit knowledge of them) and vice versa.

Dynatech is an economy simulation computer game from 1992 available through reline Software. In the Dynatech scenario, resources are exhausted, and distant planets are colonized to ensure the supply of Earth (Terra) with necessary goods. Resources such as crude oil can be conveyed and stepwisely processed in manufacturing plants to finally create end products, which are purchased by Terra. In addition spaceships are used to transport the intermediate and end products from one planet to another.

The human and computer players build up companies trying to make a fortune in this emerging market. The companies compete and cooperate in a higly dynamic market-based scenario, being able to buy or sell goods and post or accept transportation orders. In this scenario different strategies are possible. Companies may concentrate on conveying raw materials or the manufacturing of end products. Others might try to set up complete supply-chains, or assemble a fleet of spaceships focusing on the transportation domain.

In a teaching project of the VSIS group at the University of Hamburg a reimplementation of the Dynatech game was done. An initial analysis phase led to a design with six different types of agents. The manager agent is the head of a company, responsible for making all business related decisions (which orders to post or accept, which ships assign to which orders etc.). The spaceship agent represents a spaceship, that travels between planets and carries out transportation oders received from its manager. The industrial plant agent monitors the manufacturing or conveying process, notifying the manager, when incoming supply becomes too low, or the storage capacity for produced goods tends to fill up. The market agent allows transport orders and demands for goods to be posted and accepted, and assures a fair pricing. The Terra agent represents the resource and end product consumption of Earth and periodically posts corresponding demands to the market agent. Finally, a time service agent provides a global clock, and informs all agents participating in the simulation, when a new day has started.

The students dynamically formed nine groups, with six groups assigned to the identified agents. The other three groups were responsible for creating a graphical user interface, an ontology for the agent communications, and a JXTA based service discovery mechanism.

Also in a teaching project of the VSIS group at the University of Hamburg an online bookstore has been developed. The bookstore application was inspired by the book Developing Intelligent Agent Systems: A Practical Guide from Lin Padgham and Michael Winikoff (published by Wiley & Sons in 2004). The bookstore scenario is described in the book as follows:

"We would like to develop a fully online system for worldwide sale of books to customers, and a personalized, friendly user interface. The system must facilitate fast and reliable service at all stages, from locating a desired book, to delivery of the purchase. The store should have competetive prices."

The analysis phase of the project was done using the Prometheus methodology and carried further the modelling artifacts already presented in the book. As one primary result a design consisting of six different agent types was devised. The sales assistant represents a personal shopping assistant for a customer. For each customer arriving at the web site an individual sales agent is created. It has the purpose to help a customer finding and purchasing books. Therefore, it has to cooperate with external credit card or other payment companies. Customer data and offline (e.g. email) communication with a customer is handled by the customer relations agent. The delivery manager agent is resposible for a timely delivery of books to the customer. It has to organize the delivery in cooperation with certain courier and postal services. Additionally, it can be inquired about outstanding deliveries. For guaranteeing a fast delivery the bookstore stock needs to be kept up-to-date. A stock manager agent has the task to reorder book lists from a wholesale in regular time intervals and on demand. The price observer agent has the responsibility to monitor book prices in comparison to market competitors. If a book is cheaper at a competitor's site the price observer will notify the stock manager and propose a temporary price reduction. Finally, the external services also had to be represented as agent types. Adjustable bank agents as well as delivery agents have been developed.

The students worked in small groups to realize the identified agents. Other groups were responsible for creating an ontology and a data persistence layer. Additionally, by a separate group, a mobile student assistance application was developed capable of exchanging book lists (e.g. for courses) and to order subsets of the book lists at the local library or the online bookstore.