Learning Partially Observable, Deterministic Action Models (II)

We present new algorithms for learning a logical model of actions' effects and preconditions in partially observable domains. The algorithms maintain a logical representation of the set of possible action models after each observation and action execution. The algorithms perform learning in unconditional STRIPS action domains, which represent a new class of action models that can be learned tractably. Unlike previous algorithms, these algorithms are capable of learning preconditions or learning in the presence of action failures. The algorithms take time and space polynomial in the number of domain features, and can maintain a representation that stays indefinitely compact