Avances en robótica y automatización

Deep Learning and Particle Swarm Optimization for the Intelligent Rehabilitation Robot Control

Abstract

Bassant Mohammad Elbagoury

Particle Swarm Optimization (PSO) for Intelligent Control of agent autonomous rehabilitation robot is a very complex problem, especially for
stroke patients’ treatments and dealing with real-time EMG sensors readings of muscles activity states and transfer between real-time Human
motions to interface with rehabilitation robot agent or assisted-device. The field of Artificial Intelligence and neural networks plays a
critical role in modern intelligent control interfaces for robot devices. This paper presents a novel hybrid intelligent robot control that acts as
human-robot interaction, where it depends on real-time EMG sensor patients data and extracted features along with estimated knee joint angles
from Extended Kalman Filter method are used for training the intelligent controller using support vector machines trained with Adatron
Learning algorithm for handling huge data values of sensors readings. Moreover, the proposed platform for rehabilitation robot agent is tested
in the framework of the NAO Humanoid Robot agent along with Neurosolutions Toolkit and matlab code. The average overall accuracy of the
proposed intelligent motion SVM-EKF controller shows average high performance that approaches average 96% of knee motions
classifications and also good performance for comparing Extended Kalman filter knee joint angles estimations and real EMG human knee joint
angles in the framework of Human Walk Gait cycle. Also, the basic enhancement of proposing PSO optimization technique for robot knee
motion is discussed for future improvements. The overall algorithm, methodology and experiments are presented in this paper along with
future work.