Assistive robotic exoskeleton using recurrent neural networks for decision taking for the robust trajectory tracking

The development of exoskeletons has contributed to the rehabilitation of the population with different degrees of disability. These devices contemplate some feedback signals for their control; these signals can come from a user control or be taken from the brain's or muscles' user electrical response. In the specific case of exoskeletons of the lower extremities, they generally depend only on control algorithms to develop the trajectories of the user's lower extremities. In general, they also implement electromyography (EMG) interfaces as separate systems for measuring patient activity or improvements in the rehabilitation of the musculoskeletal system that carry their devices. For this reason, a human in the loop scheme was proposed in this work; a combination of a recurrent neural network (RNN) and an adaptive non-singular fast terminal sliding mode controller (ANFTSMC) strategy is employed to classify the user's movements and control the trajectories of an exoskeleton. This paper presents the construction of the electromyographic signals (EMGS) database, containing data acquired from brachii biceps and sternocleidomastoid muscles. Then, the training and validation of the RNN for the classification of EMGS. The mathematical approach of the ANFTSMC and the stability analysis of this control scheme for the trajectory tracking problem of the Sit-to-Stand task in four degrees of freedom exoskeleton. Finally, the complete human-in-the-loop system working with the RNN classifier and the ANFTSMC is established, with the advantage of being an effective, precise, and intelligent system that can be used by people with high degrees of motor disability.