The　structure　and　mathematical　modeling　of　a　Dual　-wheel　upright　self-balance　Robot　are　studied.　The　simulation　and　the　real　experiment　of　robot　are　done.　The　system　is　made　up　of　a　mobile　mechanism,　sensor　for　posture　observation　and　the　controller.　The　right　and　left　wheels　are　driven　separately　by　two　high　percision　DC　servo　motors　with　photoelectric　encoder.　Direct　current　servo　control　system　of　PWM　based　on　application　of　pulse　width　modulation　DC-DC　converters.　Gyro　scope　and　angle　sensors　are　used　for　posture　observation.　On　the　basis　of　building　up　the　system　structure　model,　the　system　mathematical　model　is　built　up　using　dynamics　mechanics　theory　and　method.　The　optimal　controller　LQR　(the　state2feed　back　controller)　is　designed　on　Matlab　the　system　has　good　robust.　By　using　TML　language　it　is　convenient　to　program　and　debug　all　kinds　of　arithmetic　on　IPM100.　The　stability　of　system　is　proved　by　simulation　results.　The　real　robot　as　human　being　can　realize　all　kinds　of　walking　control.　Validity　and　rationality　of　the　system　modeling　and　the　designed　controller　are　verified　through　the　performance　experiments　of　the　prototype.　Building　a　self-learning　robot.　©　2008　IEEE.