A Uni ed Framework of Map Learning with a Hierarchy of Probabilistic Maps

A uni ed framework of representing and acquiring maps is discussed. In the framework, maps are represented by a hierarchically structured probability distribution of sensory-motor signals, and the map learning problem is formulated as an estimation process of the distribution. Some of map learning methods proposed so far are interpreted in the general framework. A novel map learning scheme which learns multiple maps of di erent grain size concurrently to reduce the total learning cost is also proposed.