Potential role of afferent recordings as a sensory feedback in movement control systems: animal model

The authors evaluated the potential use of sensory nerve signals as sensory feedback in closed-loop control systems for FES-assisted walking. They focused on the control of flexor/extensor actions of the intact cat´s wrist, represented by the EMG signals recorded using intramuscular electrodes from two extensor muscles (Palmaris Longus and Extensor Carpi Ulnaris) and two flexor muscles (Flexor Carpi Ulnaris and Abductor Pollicis Longus) during walking on a motorized treadmill. Sensory signals were simultaneously recorded in the cat´s forelimb either from Ulnar and Median or Ulnar and Radial nerves using cuff electrodes. The use of sensory nerve signals in predicting EMG signals was seen as an important step toward designing controllers which will produce more natural movements of paralyzed limbs in people with spinal cord injuries. To extract the relationship between sensory nerve signals and EMG signals, the authors applied adaptive logic networks (ALNs), a type of artificial neural networks for supervised learning. These relationships were extracted during the training and used to predict EMG signals when only the sensory nerve signals were provided. As a measure of performance, generalization was assessed using the coefficient of determination (RSQ) for the predicted and the original EMG signals. In the present study the authors achieved very good ALN learning and generalization over the same walking session, even when they used only one of the sensory signals or data from only one gait cycle for training. Current efforts are focused on generalization over longer periods of time and between-subject generalization