Oil refineries provide the lifeblood for global economic health, and 
disruptions to their operations have major worldwide impact. For almost a 
decade, Honeywell's Automated
Reasoning Group has been working together with over a dozen major 
petrochemical companies to move their industry into the next generation of 
sophisticated software control
systems. One of the most ambitious and innovative aspects of this program has 
been the application of AI techniques to control these complex plants during 
normal, and
especially abnormal situations. 

We are developing a large-scale intelligent refinery control system to either 
assist human operators, or automate certain plant procedures directly, 
depending on what each
situation warrants. Based on reactive and procedural approaches to intelligent 
behavior, the Abnormal Event Guidance and Information System (AEGIS) will 
interact with
multiple users and thousands of refinery components to diagnose and compensate 
for unanticipated plant disruptions. Adjusting the autonomy of AEGIS's 
behavior to
intelligently adapt to each individual situation is a key requirement for 
success in the dynamic, highly-unpredictable refinery environment.

In this talk I will discuss our technical approach to the goal-setting, 
planning, and plan execution components of AEGIS, and the adjustable autonomy 
features they support.
Then, to complement this depth-first exploration of a single project, I will 
briefly provide a breadth-first overview of several other projects currently 
underway in our research
group.