This　paper　aims　to　develop　a　multi-sensor　fusion　method　for　attitude　estimation　for　a　Self-Balancing　Two-Wheeled　Robot　using　a　Micro　Inertial　Measurement　Unit.　An　adaptive　extended　Kalman　filter　is　proposed,　which　can　filter　the　random　drift　error　from　inertial　sensors,　and　also　adaptively　compensates　external　acceleration.　External　acceleration,　which　affects　attitude　estimation　based　on　accelerometers,　is　a　major　source　of　attitude　estimation　error.　The　external　acceleration　error　is　compensated　by　adjusting　the　measurement　noise　covariance　adaptively.　The　proposed　algorithm　is　verified　through　experiments.　©　2011　IEEE.