Learning of Causality by a Robot
P. Mowforth, Tatjana Zrimec · 1991
Abstract This paper describes an experiment in which a robot is allowed randomly to explore the domain of object-pushing via controlled experi ments. The experiments consist of recording signals from sensors before and after an action has taken place. Each experiment may be considered as a state transformation recorded as a sequence of attribute values. Treating each transformation as a training example, a large set of data was collected and subjected to rule induction. Robust and useful trans formations were discovered which were represented as a hierarchical qualitative model. Further, results show that an actor-oriented, co ordinate frame provides the most compact description for the problem. One final observation is that this style of experimentation offers the potential for closed-loop learning in that, unlike other domains, as far as the robot is concerned, its world is the oracle.