The GRL System: Learning Board Game Rules with Piece-Move Interactions.

Many real-world systems can be represented as formal state transition systems. The modeling process, in other words the process of constructing these systems, is a time-consuming and error-prone activity. In order to counter these difficulties, efforts have been made in various communities to learn the models from input data. One learning approach is to learn models from example transition sequences. Learning state transition systems from example transition sequences is helpful in many situations. For example, where no formal description of a transition system already exists, or when wishing to translate between different formalisms. In this work, we study the problem of learning formal models of the rules of board games, using as input only example sequences of the moves made in playing those games. Our work is distinguished from previous work in this area in that we learn the interactions between the pieces in the games. We supplement a previous game rule acquisition system by allowing pieces to be added and removed from the board during play, and using a planning domain model acquisition system to encode the relationships between the pieces that interact during a move.