Approximate Processing for Real-Time Problem-Solving in Phoenix: A beginning

摘要

This report focuses on the problem of developing real-time problem- solving capabilities which will yield the best possible solution given the available time and computational resources. The research was part of a project called Phoenix, which aims to develop intelligent real-time planners for fighting simulated forest fires. Phoenix is designed to be a testbed for experiments in distributed control, which is one of the key characteristics of battlefield planning. The report describes research focused on: a) adapting the elements of approximate processing, which have been developed for distributed knowledge-based signal interpretation, to a planning problem having striking similarities to those observed in battlefield planning, and b) developing an approach for evaluating architectures for such real-time planning systems through the use of hypothesis testing, experiment design and statistical analysis.