Joint Angle Estimation and Control of Lower Limb Exoskeleton with Surface Electromyography Inputs and Neural Networks

The Lower Limb Exoskeleton (LLE) is a powerful assistive device that has been developed for rehabilitation and movement assistance for people with lower limb disability. The actuation of LLE joints with external inputs often causes jolts and abrupt shocks to the wearer. These can be overcome by using surface electromyography (sEMG) signals from lower limb muscles of the user for actuation and this facilitates natural and energy efficient walking. In this work, an LLE with 3 degrees of freedom on each leg is considered. The hip, ankle, and knee joint reference trajectories of the LLE are generated by using a neural network trained with sEMG signals. The sEMG signals are derived from the four leg muscles viz Tibialis Anterior, Gastrocnemius Lateralis, Bicep Femoris and Vastus Lateralis. The joint angle trajectories of both right and left limbs are obtained as the output of a feed forward propagation neural network and are compared with the known reference trajectories. The control of joint angles is done by a proportional integral derivative controller, which is found to track the reference trajectories of LLE. The root mean square error and the maximum error are tabulated to measure the performance of the controller. The proposed work is compared with existing literature and found to perform better.