Abstract: Algorithms for angular acceleration estimation are investigated. Based on the analyses of recursive linear smoothed Newton predictor (RLSN) and Kalman filter (KF), a new predictive filter is proposed for acceleration estimation by combining the KF and Newton predictor (KFNP) together. This intends to achieve a wide bandwidth of estimated acceleration by reducing the phase lag introduced by filtering, and at the same time, attenuating noises. Extensive experiments are conducted on the first joint of 2-DOF direct-drive manipulator. The frequency responses of the acceleration are estimated by RLSN, KF, as well as KFNP are measured and respectively compared with the one of the accelerometer output. The results demonstrate the consistency between estimated and measured acceleration, and further reveal the possibility of utilizing estimated acceleration for acceleration feedback control instead of a measure one.

Key words: Angular acceleration estimation, Kalman filter, Newton predictor