Model predictive current control can achieve fast dynamic response and satisfactory steady-state performance for induction motor (IM) drives. However, many motor parameters are required to implement the control algorithm. Consequently, if the motor parameters used in the controller are not accurate, the performance may deteriorate. In this paper, a new robust predictive current control is proposed to improve robustness against parameter mismatches. The proposed method employs an ultra-local model to replace the mathematical model of the IM. Additionally, to improve the control performance, a linear extended state observer is developed for disturbance estimation. Experimental tests confirm that satisfactory tracking performance can still be obtained although the motor parameters may not be accurately set in the controller.