The CWave 2000 Visual Agent Workbench

Dissertation Abstract

Over the last several years, measurement technology has undergone a transformation from systems with many transducers attached to a central computer to distributed measurement systems where each transducer has an attached CPU, downloadable code, and a network connection.   Even though measurement technology has changed dramatically, measurement systems are still built using old-fashioned and difficult to debug program logic control (PLC) technology, which lacks important features such as fault tolerance, flexibility, and visualization capabilities.

Research in software agent technology has been underway for several years, resulting in many high-performance agent systems. With a few exceptions, most existing agent systems focus on low-level technical details, such as performance, mobility and communication and do not address deployment, scaling, and especially the management issues of hundreds or thousands of agents.  Many of them also ignore higher-level issues, such as intelligence and autonomous behavior.

Based on experiments and case studies in two different domains, we demonstrate that software agents combined with a visual programming lanuage address the shortcomings of the PLC technology mentioned above. In particular, agent autonomy and hierarchical structuring reduce communication overhead and permit redundancy to be built into the system.  While a small-scale, hierarchically structured distributed measurement system (DMS) can be managed without a graphical environment, larger-scale systems benefit from visual metaphors to effectively manage an arbitrary number of measurement nodes.

In support of this claim, we have constructed a prototype of a visual agent workbench that can be used to design, simulate and manage agent systems for distributed measurement and control applications. Instead of adding visualization and management capabilities to an existing agent architecture, we have taken the unique approach of combining a visual programming language with our own agent architecture customized for DMS applications. We show with two case studies that the resulting system, a combination of a highly customizable, general-purpose visual programming environment with an extensible agent architecture, can be used both as a rapid prototyping tool, and also as a development tool for DMS systems. The case studies demonstrate the deployment of a large number of agents over a local area network and use of the workbench to construct and simulate a distributed measurement application with built-in redundancy. We also show the benefits of visually constructing loggers, filters, and performance measurement tools and finally compare the benefits of the visual metaphor with conventional systems.

Download this dissertation in PDF format.