"New Directions in AI Planning: High-Stake Planning and Optimal Replanning"

2/13/2004: 10:30am - 12:00pm
11th FL Large Conference Room

Abstract: Autonomous agents must be able to make good decisions in complex situations that involve a substantial degree of uncertainty, yet find solutions in a timely manner despite the large number of potential contingencies. Examples include decision-support systems and mobile robots. In this talk, I will cover two aspects of this problem, namely planning in high-stake decision situations and fast replanning in highly dynamic decision situations. Both of these case studies have in common that the resulting methods combine ideas from artificial intelligence with ideas from other decision-making disciplines. Autonomous agents often plan in high-stake decision situations, that is, decision situations that potentially result in the loss of large amounts of money. I will discuss how to build artificial intelligence planners that fit the preference models of human decision makers better than current planners, by combining constructive methods from artificial intelligence with more descriptive methods from utility theory. I will then show how to apply the resulting theory to auction planning for supply-chain management systems. Similarly, autonomous agents often operate in domains that are only incompletely known or change dynamically. In this case, they need to be able to replan quickly as their knowledge changes. Since replanning from scratch is often time consuming, it can be advantageous for them to use planning methods that modify the previous plans or plan-construction processes locally. I will discuss replanning methods that combine ideas from artificial intelligence and algorithm theory. I will then show how to apply the replanning methods to STRIPS-type planning and navigation planning for mobile robots, although it is not yet understood when they are more efficient than conventional planning methods. This is joint work with Colin Bauer, David Furcy, Richard Goodwin, Maxim Likhachev, and Yaxin Liu.

About Sven Koenig: Sven Koenig is interested in planning and learning. He received his Ph.D. from Carnegie Mellon University in 1997. He also received an NSF CAREER award and an IBM Faculty Partnership Award, which funded some of the research reported on in this talk.

Last updated: Mon Jun 19 17:44:06 2006