A Logic-based Approach to Dynamic Programming

Steffen Hölldobler and Olga Skvortsova

We present a first-order value iteration algorithm that addresses the scalability problem of classical dynamic programming techniques by logically partitioning the state space. An MDP is represented in the Probabilistic Fluent Calculus, that is a first-order language for reasoning about actions. Moreover, we develop a normalization algorithm that discovers and prunes redundant states. We have implemented our approach and describe some experimental results.

This page is copyrighted by AAAI. All rights reserved. Your use of this site constitutes acceptance of all of AAAI's terms and conditions and privacy policy.