Abstract: Due to the problem that the operating state parameters of the robot system can't be predicted and estimated entirely by the offline identification mode， we designed an adaptive unscented Kalman filter algorithm based on MIT rules. This method took the difference between the actual value and the estimated value of information variance as the index parameter and updated the unknown parameters by the gradient descent method to realize adaptive control. Theoretical analysis and experimental results showed that when the statistical noise characteristics of the system changed，the proposed adaptive filter algorithm could automatically adjust its parameters to reduce the influence of the system's prior noise information on the filter performance. This method effectively improves the stability of the filter and the accuracy of the estimation.

Key words: adaptive algorithm, unscented Kalman filter, robot, online estimation, state parameters