Neuro-Identifier for the Musculoskeletal System of the Upper-Limb to Map Electromyographic Signals to Inverse Kinematics

This paper presents a non-parametric mathematical model based on differential neural networks (DNNs) for the upper limb's neuromusculoskeletal system. The U-LIMB dataset is the source of input-output pairs that relate electromyographic (EMG) signals from different arm muscles with daily patient movements. This information trains the free weights of the DNN through Lyapunov stable learning laws. With the model provided by the DNN, it is possible to get online the inverse kinematic information that defines the desired trajectories for a virtual designed arm with five degrees of freedom. Simulation results demonstrate the application of these trajectories by implementing a set of proportional derivative controllers that ensured the correct movement of the virtual arm in response to a given set of EMG signals.