Trajectory tracking in upper limb rehabilitation exercises is utilized for repeatability of joint movement to improve the patient�s recovery in the early stages of rehabilitation. In this article, non-linear active disturbance rejection control as a combination of non-linear extended-state observer and non-linear state error feedback is used for the sinusoidal trajectory tracking control of the two-link model of an upper limb rehabilitation exoskeleton. The two links represent movements like flexion/extension for both the shoulder joint and the elbow joint in the sagittal plane. The Euler�Lagrange method was employed to acquire a dynamic model of an upper limb rehabilitation exoskeleton. To examine the efficacy and robustness of the proposed method, four disturbances cases in simulation studies with 20 parameter variation were applied. It was found that the non-linear active disturbance rejection control is robust against disturbances and achieves better tracking as compared to proportional�integral�derivative and existing conventional active disturbance rejection control method. © IMechE 2020.