Abstract: For the problem that the extended Kalman filter(EKF) is difficult to design and prone to diverge,the unscented Kalman filter(UKF) algorithm based method is presented to solve the problems of the filter design and convergence.The error from the low-cost inertial gyro and accelerometer is compensated to achieve optimal attitude estimation.The filtered model is applied to the two-wheeled self-balanced robot system.Experimental results demonstrate that for the UKF,the parameter design is easier,the attitude estimation error is smaller,the covariance estimation is better than those of the EKF,while the estimation precision and the computational costs are comparable.Consequently,the UKF is suitable for the real-time attitude estimation of the two-wheeled self-balanced robot in the fast and maneuverable process.