This　paper　presents　an　Artificial　Brain　System　of　a　Mazelike　robot,　which　comprises　of　a　perception　unit　and　decision-making　unit.　The　perception　module　is　based　on　ARTl　neural　network,　trained　to　identify　the　signposts　of　the　maze;　decision-making　units　is　based　on　behavior　probability　matrix　P,　and　it　uses　reinforcement　learning　to　update　the　action　strategy.　The　maze　in　which　the　robot　would　navigate　has　signposts　at　every　intersection.　The　signposts　are　2-D　symbols　with　noise.　In　the　simulation　tests,　the　robot　moves　randomly　in　the　maze.　By　adjusting　the　vigilance　parameter　P　and　reinforcement　constant　CRF,　the　robot　will　eventually　pass　through　the　maze　after　a　learning　process　during　the　self-exploration.　The　simulation　show　that　the　artificial　brain　system　can　be　self-organized　to　make　sense　of　the　signposts　and　successfully　guide　the　robot　through　the　maze.　©　2009　IEEE.