On the automatic Entropy-based construction of Probabilistic Automata in a Learning Robotic Scenario

When a robot interacts with the environment producing changes through its own actions, it should find opportunities for learning and updating its own models of the environment. A robot that is able to construct discrete models of the underlying dynamical system which emerges from this interaction can guide its own behavior and adapt it based on feedback from the environment. Thus, the induction of probabilistic automata from this sensorimotor loop might be useful for planning/learning tasks. These probabilistic automata can be used as a prediction tool, as a means to assess the uncertainty or predictability of specific action consequences and thus, as a tool for an active learning method.