Coordinating Intelligent Agents

 

Abstract

This talk will focus on how to get organizations---multiple software agents and humans---to coordinate their activities when they are working on shared, loosely coupled problems, such as engineering design or information gathering. I'll describe some useful representations (including TAEMS [Task Analysis and Environment Modeling System]) for annotating an agent's representation of its activities, some approaches (including GPGP [Generalized Partial Global Planning]) to designing coordination mechanisms that are adapted to some particular problem-solving environment, and an agent toolkit (DECAF [Distributed Environment-Centered Agent Framework]) encompassing these ideas. DECAF combines features of my RETSINA work at CMU and my work on coordination at UMass. DECAF uses Java threads to allow an agent to communicate, plan, schedule, and execute domain actions concurrently, and focuses on providing agent-level programming tools (graphical plan editor, pre-written middle-agents) for rapid construction and experimentation with multiple agents over any TCP/IP based network. Examples will be drawn from various projects in distributed hospital patient scheduling and distributed information gathering, including our recent work in automating genomic annotation (BioMAS).

 

BACK