There seems to be no end to what the definition of an agent or a multi-agent system is and I think the answer to this is not important. I see multi-agent system as a new paradigm for building distributed applications. Just like OOP Methodologies, multi-agent methodology would be one which provides a powerful tool to visualize and build distributed applications. I had the oppurtunity to work on a couple of agent related projects in the LAIT lab. My long term goal is to convince people to use semantic web languages and see its use in information retrieval, knowledge management and as an instrument for knowledge dissemination.
DQL defines a set of protocol elements that enables an agent to query a KB. These protocol elements are described in DAML+OILHowever, DQL does not specify the interaction protocol between agents. It turns out that some FIPA communicative acts and interaction protocols can used with DQL. We implemented DQL with interaction protocols on ITTalks. ITTalks is a talk notification service with human and agent interfaces to it.
According to me, planning is a very natural solution for service composition. However there are several issues that need to be resolved before this concept can be adopted. Inspite of the interest shown by the industry, use of web services for building distributed applications is looked upon as a research endeavor. DAML-S is one of the first applications of DAML+OIL. DAML-S provides an ontology for describing web services. DAML-S ontologies cannot (are not designed to) capture domain knowledge. We believe, to compose non-trivial services, domain knowledge needs to be plugged-in somewhere. DAML+OIL provides the neccessary constructs to capture this additional knowledge. We have a built a backward chaining planner in JESS that composes complex services from atomic services.
Hybrid Architecture is a relatively new paradigm for building robot control systems. However this involves the fusion of two very different architectures: Deliberative and Reactive Architecture. Interfacing these two is a very interesting problem. We proposed a methodology for building hybrid architectures primarily using the persistence of data as a guiding factor. We also built a prototype that involved a herding scenario. Teambots is the simulator we used to build our scenario and the robots. This is a Java based simulator and very easy to work with. If someone is interested in this work and would like to extend on this, I am more than willing to provide the source code for the hybrid architecture.
Mithun Sheshagiri, Marie desJardins, Tim Finin, A Planner for Composing Service Described in DAML-S, Workshop on Planning for Web Services, International Conference on Automated Planning and Scheduling, Trento, July 2003. download
Anugeetha Kunjithapatham, Mithun Sheshagiri, Tim Finin, Anupam Joshi, and Yun Peng, Personal Agents on the Semantic Web, submitted to the Second International Semantic Web Conference (ISWC2003).
Subhash Kumar, Anugeetha Kunjithapatham, Mithun Sheshagiri, Tim Finin, Anupam Joshi, Yun Peng, and R. Scott Cost, A personal agent application for the semantic web, AAAI Fall Symposium on Personalized Agents, North Falmouth, Nov 15-17, 2002.download
Mithun Sheshagiri and Marie desJardins, Data Persistence: A Design Principle for Hybrid Robot Control Architectures, Paper, International Conference On Knowledge Based Computer Systems, Mumbai, India 2002. download