The increasing number of elderly in the world reveals today new societal challenges, particularly in terms of healthcare and assistance services. With recent advances in technology, robotics appears as a promising solution to develop systems that improve the living conditions of this aging population. This thesis aims at proposing and validating an approach for robust control of an active orthosis, based on the subject intention. This orthosis is designed to assist flexion/ extension movements of the knee for people suffering from knee joint deficiencies. The proposed second order sliding mode control allows to take into account the nonlinearities and parametric uncertainties resulting from the dynamics of the equivalent lower limb-orthosis system. It also ensures on one hand, a good tracking performance of the desired trajectory imposed by the therapist or the subject itself, and on the second hand, a satisfactory robustness with respect to external disturbances that may occur during flexion and extension of the knee joint. In this thesis, a neural model based on Multi-Layer Perceptron is used to estimate the subject's intention from the measurement of the EMG signals characterizing the voluntary activities of the quadriceps muscle group. This approach overcomes the complex modeling of the muscular activation and contraction dynamics. All the proposed approaches in this thesis have been validated experimentally with the voluntary participation of several healthy subjects
