Learning action models from plan examples with incomplete knowledge

AI planning requires the definition of an action model using a language such as PDDL as input. However, building an action model from scratch is a difficult and time-consuming task even for experts. In this paper, we develop an algorithm called ARMS for automatically discovering action models from a set of successful plan examples. Unlike the previous work in action-model learning, we do not assume complete knowledge of states in the middle of the example plans; that is, we assume that no intermediate states are given. This requirement is motivated by a variety of applications, including object tracking and plan monitoring where the knowledge about intermediate states is either minimal or unavailable to the observing agent. In a real world application, the cost is prohibitively high in labelling the training examples by manually annotating every state in a plan example from snapshots of an environment. To learn action models, our ARMS algorithm gathers knowledge on the statistical distribution of frequent sets of actions in the example plans. It then builds a propositional satisfiability (SAT) problem and solves it using a SAT solver. We lay the theoretical foundations of the learning problem and evaluate the effectiveness of ARMS empirically.