Map Learning with Uninterpreted Sensors and Effectors

This dissertation presents a set of methods by which a learning agent, called a \critter, " can learn a sequence of increasingly abstract and powerful interfaces to control a robot whose sensorimotor apparatus and environment are initially unknown. The result of the learning is a rich, hierarchical model of the robot's world (its sensorimotor apparatus and environment). The learning methods rely on generic properties of the robot's world such as almost-everywhere smooth e ects of actions on sensory features. At the lowest level of the hierarchy, the critter analyzes the e ects of its actions in order to de ne control signals, one for each of the robot's degrees of freedom. It uses a generate-andtest approach to de ne sensory features that capture important aspects of the environment. It uses linear regression to learn action models that characterize context-dependent e ects of the control signals on the learned features. It uses these models to de ne high-level control laws for nding and following paths de ned using constraints on the learned features. The critter abstracts these control laws, which interact with the continuous environment, to a nite set of actions that implement discrete state transitions. At this point, the critter has abstracted the robot's world to a nite-state machine and can use existing methods to learn its structure. ii