CSAIL Event Calendar: Previous Series
Planning and Programming with First-order Representations of Markov Decision Processes
Speaker: Craig Boutilier , Department of Computer Science, University of Toronto
Two paradigms have emerged for the control of intelligent agents, planning and programming, each with each own benefits and drawbacks. We propose a framework for agent programming which allows the seamless integration of explicit agent programming with decision-theoretic planning. Specifically, the DTGolog model allows one to partially specify a control program in a high-level, logical language, and provides an interpreter that, given a first-order logical axiomatization of a domain, will determine the optimal completion of that program (viewed as a Markov decision process). We demonstrate the utility of this model with results obtained in an office delivery robotics domain. We also make a number of suggestions for improving the computational efficiency of the existing interpreter, and extensions that relax some of our underlying assumptions. Time permitting I will describe recent developments on the application of first-order regression to the problem of dynamic programming in first-order MDPs. Our first-order decision-theoretic regression algorithm allows the construction of optimal policies for such MDPs without explicit state space enumeration.