Thursday, September 11, 2008

Planning & Diagnosis

Hey Everybody,

I'm very interested in a notion of self-awareness for systems. In the real world it is necessary for systems to understand their own conditions and implications on their capabilities. I have been looking into the integration of planning and diagnosis which we call Pervasive Diagnosis over the last year. In this approach a model-based planner uses a system description to create a plan that achieves production goals and the same model is used by a model-based diagnosis to indirectly infer the condition of components in a system from partially informative sensors. Pervasive Diagnosis opens up new opportunities to efficiently exploit diagnostic information for the optimization of model-based systems. Hard to diagnose intermittent faults or shifting capabilities which would have required expensive stoppages can now be addressed on line during planning. While pervasive diagnosis has interesting theoretical advantages, we have shown that a combination of heuristic planning and classical diagnosis can be used to create practical real time applications as well.

I would highly appreciate comments and feedback. Thanks!! :-luk


TheKing said...
This comment has been removed by the author.
TheKing said...

what do you actually mean by "model-based" planner? are you talking about the PDDL description of the world as the model or something else. Indeed it is very interesting topic. Here at Canberra we do have guys who do diagnosis and also have guys who are doing planning. Priscilla, one of my colleague, was very interested in this topic. She is going to ICAPS. May be we can catch-up.

lumble said...

By model-based I mean that the planning system does it's decisions based on a pre-descripted model such as a model in PDDL. Other than approaches where the planner learns the model as part of the planning problem. I would love to catch-up, but first I have to catch some sleep - it is 1 am. Cu tomorrow.