Model-free based adaptive finite time control with multilayer perceptron neural network estimation for a 10 DOF lower limb exoskeleton

This article presents a Model-Free Adaptive Nonsingular Fast Terminal Sliding Mode Controller with Super Twisting and Multi-Layer Perceptron (MLP) neural network for motion control of a 10 DOFs lower limb exoskeleton used in rehabilitation. The proposed controller employs a second-order ultra-local model to replace the complex dynamics of the exoskeleton and uses an MLP neural network to estimate the lumped disturbance of the ultra-local model. To ensure accurate tracking of the desired trajectory and address the estimation errors of the MLP, an Adaptive Nonsingular Fast Terminal Sliding Mode Controller is introduced. Moreover, a Super Twisting approach is employed to eliminate the chattering phenomenon. The system's stability is analyzed using Lyapunov theory, and the desired trajectories are obtained from surface electromyography (EMG) signal measurements. The effectiveness of the proposed controller is validated through co-simulation experiments using SolidWorks, Simscape Multibody, and MATLAB/Robotics Toolbox. Results demonstrate significant improvements in stability and precision compared to existing model-free controllers.