For patients with stroke, hemiplegia and ankle injuries, ankle rehabilitation robots can provide personalized rehabilitation treatment programs through quantitative evaluation and precise control. In this paper, based on differential evolution (DE) optimization algorithm and fuzzy nonlinear auto regressive with exogenous inputs (NARX) model, an iterative learning model predictive controller is constructed to achieve accurate and robust trajectory tracking control of an ankle rehabilitation robot driven by pneumatic muscle. Firstly, the T-S fuzzy NARX model is applied to characterize the forward and inverse hysteresis process of pneumatic muscle actuator (PMA) in this ankle rehabilitation robots and differential evolution optimization (DE) algorithm realize high precision parameters identification. Then, the optimization of objective function is compared with genetic algorithm and particle swarm optimization. Secondly, the iterative learning model predictive control (ILMPC) controller is designed and the trajectory tracking control scheme of an ankle rehabilitation robot is presented based on the ILMPC controller. Finally, the experiments with different trajectories and subjects are carried out to verify the control performance of the proposed control method. The experimental results indicate that the proposed ILMPC controller can converge normally and has good control performance with different operating trajectories and subjects.