Feedback-error learning control for powered assistive devices

Dissertação de mestrado integrado em Engenharia Eletrónica Industrial de Computadores ; Gait pathologies often produce abnormal gait patterns, affecting human mobility. Powered assistive devices, such as lower-limb exoskeletons and orthoses, are starting to complement gait rehabilitation, to actively aid or restore the abnormal gait pattern. The human motor control system starts to influence the design of bioinspired architectures for these devices, comprising the definition of distinct levels of controllers (high-, mid-, and low-level) distributed hierarchically. Low-level controllers play an important role in this architecture, ensuring time-effective assistance adaptive to user’s needs as gait speed and trajectory. The main goal with this dissertation is the development of a real-time Feedback-Error Learning (FEL) low-level control to be integrated into a bioinspired control architecture approached in a Stand-alone, Active Orthotic System - SmartOs. The FEL control was performed by means of an Artificial Neural Network (ANN) as a feedforward controller to acquire the inverse model of the assistive device, and a Proportional-Integral-Derivative (PID) feedback controller to guarantee stability and handle with disturbances. A Powered Knee Orthosis and Powered Ankle-Foot Orthosis were used as the assistive devices and a positionbased tracking assistive strategy was applied. A validation without human load and with two subjects walking in a treadmill at 0.8, 1.0 and 1.2 km/h with the two assistive devices, controlled by the Feedback-Error Learning control, was performed. The ANN took around 90 s to learn the inverse model of the assistive device, demonstrating versatility and steadiness when changes to the magnitude and speed of the input trajectory were applied. The feedback controller guaranteed stability and shown good reactions to the applied disturbances. The implemented FEL control was capable to decrease the angular position error by 15% and to eliminate 0.25 s of phase delay when compared to a solo PID ...