Aiming　at　the　error　from　the　accelerometer　(MMA7361)　and　the　overshoot　from　the　gyroscope　(ENC03)　used　in　the　small　two-wheeled　self-balanced　robot　attitude　measuring,　a　method　combining　hardware　with　software　was　proposed　to　get　a　more　accurate　attitude,　that　is　to　implement　the　information　fusion　of　accelerometer　and　gyroscope　by　Kalman　filter　on　the　basis　of　the　improvement　of　sensor　hardware　circuit.　The　attitude　measuring　experimental　results　show　that　the　method　combining　software　with　hardware　for　the　robot　attitude　information　optimal　estimation　is　feasible　and　effective,　and　the　small　two-wheeled　robot　self-balancing　control　experiment　further　illustrates　that　the　attitude　data　is　accurate　and　reliable.