Figure 1

Experimental setup. Surface EMG signals from 12 arm muscles, kinematic data from electromagnetic sensors placed on the hand and shoulder, and grip force exerted by the thumb were recorded while healthy individuals executed random, unrestricted movements without (unloaded) or with (loaded) objects in their hand. These data served as inputs to train artificial neural networks. Once trained, a new set of kinematic and grip force data served as inputs to the artificial neural network in order to predict the associated patterns of EMG activity across the 12 muscles (Lats: latissimus dorsi, PecMaj: pectoralis major, Teres: teres major, SerAnt: serratus anterior, AntDelt: anterior deltoid, PostDelt: posterior deltoid, Tricep: triceps brachii, Bicep: biceps brachii, Brach: brachialis, BrachRad: brachioradialis, ECR: extensor carpi radialis, FCR: flexor carpi radialis) (adapted from Johnson and Fuglevand, 2009).