A　direct　adaptive　controller　based　on　improved　radial　basis　function　(RBF)　neural　networks　(NN)　is　proposed　for　an　omni-directional　mobile　robot　(OMR).　The　OMR　is　a　multi-input　and　multi-output　(MIMO),　unmodeled　and　uncertain　nonlinear　system　which　is　difficult　to　be　modeled　due　to　a　large　number　of　immeasurable　and　uncertain　variables.　To　model　the　system　exactly　and　increase　the　real-time　performance,　a　novel　direct　adaptive　control　approach　based　on　improved　RBF-NN　is　designed　to　approximate　the　OMR,　which　needs　no　explicit　knowledge　of　the　uncertain　nonlinear　MIMO　system.　Besides　the　kinematics,　the　dynamics　of　the　OMR　are　considered　to　perform　tasks　with　heavy　load　transportations　or　high　speed　movements.　A　stable　on-line　adaptive　law　is　derived　and　proved　using　Lyapunov　stability　theory.　The　proposed　controller　is　applied　the　OMR　trajectory　tracking　and　shows　excellent　robustness　and　stability.　The　simulation　results　demonstrate　the　feasibility　and　validity　of　proposed　scheme.