This paper presents methods of combining the extended Kalman filtering/smoothing, the maximum likelihood (ML) estimation, and expectation maximization (EM) for estimating the rotor speed and flux of the induction motor online in the case in which process and measurement noise statistics are not known.