Abstract:In order to realize the robot kinematic parameter calibration, a method is proposed to acquire its real time terminal attitude information using inertial measurement unit (IMU). However, during the robot dynamic attitude measurement with IMU, there exist some problems that affect the accuracy of measurement, such as the superposition of harmful accelerations (other acceleration information except gravity acceleration signal) in the accelerometer signal, the difficulty in obtaining the statistical characteristic parameters of noise and the drift of gyroscope signal with time. Aiming at these problems, an improved algorithm of adaptive extended kalman filter (EKF) for attitude measurement is designed. Based on the EKF model, this algorithm firstly constructs the first level measurement noise variance matrix, sets up weighted factors and reduces the influence of harmful acceleration signals on measurement result. Secondly, the idea of fading memory factor is introduced in SageHusa adaptive filter algorithm to track the measurement noise of sampling data in real time and construct the second level measurement noise variance matrix. Finally, the quaternion algorithm of attitude updating is adopted to conduct data fusion and correct the error caused by gyroscope signal drift. The experiment results show that compared with the SageHusa adaptive filter algorithm, the improved selfadaptive EKF algorithm can effectively improve the attitude measurement accuracy; the mean absolute errors of pitch and roll at peak height are reduced by 50% and 3643%, respectively, and the mean absolute errors of pitch and roll at peak valley are reduced by 1428% and 1944%, respectively.