Abstract: The research on two-wheeled self-balancing robot has gained momentum over the last decade at research, industrial and hobby level around the world. This paper deals with the modeling of two-wheeled self-balancing robot. This paper proposes a new method by which the error of two self-balancing robot sensors can be reduced and avoids traditional Kalman filter which cannot meet real-time modulation. In this paper, Correction algorithm can come out real-time robot posture in the right way according to the characteristics of navigation sensor error from the iteration of nonlinear least-squares error model based on the method Dynamic Regulator. By computer simulation, an error through the gyro and accelerometer can be corrected. Kalman filter fused the data of gyroscope and accelerometer adaptation and errors of the sensors pose estimation can be corrected.