Toward Learning the Causal Layer of the Spatial Semantic Hierarchy Using SOMs

The Spatial Semaqtic Hierarchy (SSH) is a multi-level representation of the cognitive map used tbr navigation in largescale space. We propose a method for learning a portion of this representation, specifically, the representation of views in the causal level of the SSH using sell-organizing neural networks (SOMs). We describe the criteria that a good view rep- ,'eseqtation should meet, and why SOMs are a promising view representation. Our preliminary experimental results indicate that SOMs show promise as a view representation, though there are still some problems to be resolved.