Symbolic Dynamic Programming
From International Center for Computational Logic
Symbolic Dynamic Programming
project thesis by Olga Skvortsova
- Supervisor -
- Wissensverarbeitung
- Start 2002
- Download
A symbolic dynamic programming approach for solving first-order Markov
decision processes within the situation calculus is presented. As an alternative specification language for dynamic worlds the fluent calculus is chosen and the fluent calculus formalization of the symbolic dynamic programming approach is provided. The major constructs of Markov decision processes such as the optimal value function and the policy are logically represented. The technique produces a set of first-order formulae that minimally partitions the state space. Consequently, the symbolic dynamic programming algorithm presented here does not require neither state nor action space enumeration, thereby solving the drawback of classical dynamic programming methods.
