A Biologically Inspired Spatial Computer That Learns to See and Act

摘要

Vision and motor control are usually studied as separate phenomenon. They perform very different functions, they are performed by different regions of the brain, and one is perception while the other is actuation. The two structures did, however, co-evolve. While they are different structures they work together in reasoning about and manipulating the outside world. Both structures have some similar attributes. For example both the motor cortex and the visual cortex are laid out in a manner that preserves topological adjacency and the hippocampus, where positional awareness is represented also represents places in the world through a topological map. In all case the layout of the areas suggests algorithms that depend upon propagation through a kind of spatial computer in order to solve navigational tasks that combine perception, actuation, and spatial awareness. In this paper we take the position that it makes sense to study the computational aspects of learning to perform such tasks together rather than as separate disciplines and that by observing the similarities of the layouts of the associated areas we can gain some insight into a general learning engine that utilizes spatial computing principles in order to achieve complex behaviors in a complex world that can only be modeled imprecisely. This paper describes such an approach embedded within simple robotic devices.