Eventos Anais de eventos
ENEBI 2018
6º Encontro Nacional de Engenharia Biomecânica
PATIENT TORQUE ESTIMATION FOR OPTIMAL CONTROL OF A ROBOTIC REHABILITATION DEVICE
Submission Author:
Leonardo José Consoni , SP
Co-Authors:
Leonardo José Consoni, Guido Gómez, Adriano Siqueira
Presenter: Leonardo José Consoni
doi://10.26678/ABCM.ENEBI2018.EEB18-0181
Abstract
Robotic rehabilitation might be a solution for the growing number of people with neuromotor disability. When programming robust control for therapy robots, the large variability in human body dynamics leads developers to leave it as unmodeled uncertainty. Here we propose and test a way to insert patient force estimates in a LQG compensator loop. Identification of a linearized leg model is combined with a Kalman observer to get a full LQG knee movement controller. A multilayer perceptron neural network is trained from reference trajectories and inverse dynamics data to calculate human-generate torque, subtracted from LQG state feedback, so that the controller is more compliant and incites orthosis user activity. Preliminary simulations of an impaired patient are used to validate the complete system, and to compare it to traditional approach. In the end, the more interesting results are discussed and considerations about shortcomings and future developments are made
Keywords
Rehabilitation Robotics, Optimal Control, Artificial neural networks, Kalman Filter, System Identification
