Markov Decision Processes Specified by Probabilistic Logic Programming: Representation and Solution

Abstract

Probabilistic logic programming combines logic and probability, so as to obtain a rich modeling language. In this work, we extend PROBLOG, a popular probabilistic logic programming language, with new constructs that allow the representation of (infinite-horizon) Markov decision processes. This new language can represent relational statements, including symmetric and transitive definitions, an advantage over other planning domain languages such as RDDL. We show how to exploit the logic structure in the language to perform Value Iteration. Preliminary experiments demonstrate the effectiveness of our framework.