Agent Based Architectures for Dynamic Crisis Management

摘要

The focus of this research was on how human users can connect to, participate in, and control agent-based systems. One of the primary accomplishments was the development of a collaborative, dialogue-based problem solving model. The approach incorporates mixed-initiative, dialogue and agent- based technologies. The model was built in a fashion where the human works with an automated intelligent assistant agent that coordinates the interaction with other agents. The model provides a framework for interpreting the user's intentions in the interaction. A dialogue-based system approach proved to be an effective way to support humans when managing agents. Such an approach facilitates intuitive and natural human-computer interaction. The underlying problem solving model supports domain-independent collaboration with a human user. The system was built off of a previously successful dialogue-based mixed- initiative planning system called TRIPS (The Rochester Interactive Planning System). This report discusses the architecture and infrastructure for the system developed. Detail is provided on the model of collaborative problem solving. A number of experiments involving integrated, end-to-end human in the loop agent-based systems are also discussed.