Multi-layered learning systems for vision-based behavior acquisition of a real mobile robot

This paper presents a series of the studies of decomposing the large state/action space at the bottom level into several subspaces and merging those subspaces at the higher level. This allows the system to maintain computational resources assigned to the modules compact and small, to reuse the policies learned before, and therefore to avoid the curse of dimension. To show the validity of the proposed methods, we apply them to a simple soccer situation in the context of RoboCup, and show the experimental results.