Learning Generalized Policy Automata for Relational Stochastic Shortest Path Problems

Several goal-oriented problems in the real-world can be naturally expressed as Stochastic Shortest Path Problems (SSPs). However, the computational complexity of solving SSPs makes finding solutions to even moderately sized problems intractable. Currently, existing state-of-the-art planners and heuristics often fail to exploit knowledge learned from solving other instances. This paper presents an approach for learning Generalized Policy Automata (GPA): non-deterministic partial policies that can be used to catalyze the solution process. GPAs are learned using relational, feature-based abstractions, which makes them applicable on broad classes of related problems with different object names and quantities. Theoretical analysis of this approach shows that it guarantees completeness and hierarchical optimality. Empirical analysis shows that this approach effectively learns broadly applicable policy knowledge in a few-shot fashion and significantly outperforms state-of-the-art SSP solvers on test problems whose object counts are far greater than those used during training. Running example: The planetary rover example can be expressed using a domain that consists of parameterized predicates connected(l x , l y ), in-rover(r x ), rock-at(r x , l x ), and actions load(r x , l x ), unload(r x , l x ), and move(l x , l y ). Object types can be denoted using unary predicates location(l x ) and rock(r x ). l x , l y , and r x are parameters that can be instantiated with different locations and