Feb 20, 2007: Automated Planning for a Structured, Uncertain World
Filed in: Colloquium
Daniel Bryce, Arizona State University
Plan synthesis is a problem in Artificial Intelligence that has been studied by both logicians and decision theorists. Logicians perfect deterministic models to capture and exploit problem structure, leading to the scalability of many classical planners. Decision theorists advocate stochastic models, such as Markov decision processes, which despite their expressiveness remain conspicuously inefficient. In this talk, I will discuss my efforts toward combining the strengths of logical and probabilistic reasoning techniques for scaling up conditional (contingency) plan synthesis. Specifically, I will describe the adaptation of a popular data structure, called the planning graph, used for heuristic search guidance, from its origins in classical (deterministic) planning to deal with partially observable states and stochastic actions.
Daniel Bryce is a Ph.D. Candidate in the Computer Science Department at Arizona State University, where he is a member of the Yochan Research Group. His research interests encompass plan life-cycle issues, such as synthesis, execution, explanation, and modeling. He is also interested in applications of planning to problems facing systems biology, manufacturing, space exploration, and robotics. His work has been published at AAAI, ICAPS, IJCAI, UAI, AI Magazine and JAIR. He also co-delivered tutorials on these topics at ICAPS and IJCAI.
Abstract
Plan synthesis is a problem in Artificial Intelligence that has been studied by both logicians and decision theorists. Logicians perfect deterministic models to capture and exploit problem structure, leading to the scalability of many classical planners. Decision theorists advocate stochastic models, such as Markov decision processes, which despite their expressiveness remain conspicuously inefficient. In this talk, I will discuss my efforts toward combining the strengths of logical and probabilistic reasoning techniques for scaling up conditional (contingency) plan synthesis. Specifically, I will describe the adaptation of a popular data structure, called the planning graph, used for heuristic search guidance, from its origins in classical (deterministic) planning to deal with partially observable states and stochastic actions.
Bio
Daniel Bryce is a Ph.D. Candidate in the Computer Science Department at Arizona State University, where he is a member of the Yochan Research Group. His research interests encompass plan life-cycle issues, such as synthesis, execution, explanation, and modeling. He is also interested in applications of planning to problems facing systems biology, manufacturing, space exploration, and robotics. His work has been published at AAAI, ICAPS, IJCAI, UAI, AI Magazine and JAIR. He also co-delivered tutorials on these topics at ICAPS and IJCAI.