Estimation of desired motion intention using extreme learning machine for upper limb assist exoskeleton

摘要

One of the complicated issue in compliance control for rehabilitation and assistive robots is to predict right human's motion intention. In this paper, we have proposed an algorithm to estimate Desired Motion Intention (DMI) so that better compliance could be provided by the rehabilitation and assitive robots. Proposed algorithms is based on Extreme Learning Machine (ELM) and takes inputs from different sensors. These sensors provide information about current angular position, speed and the force applied by the human on robot. Proposed algorithm is free from the issues which appear in traditional Radial Basis Function Neural Network (RBFNN) such as local minima, selection of suitable parameters, slow convergence of adaptation law and over-fitting. These issues cause many problems for such algorithm in tuning for each different individual and makes it impractical for our application. Developed algorithm is experimentally evaluated for two kinds of trajectories which are employed for the Activities of Daily Living (ADL) for rehabilitation purposes. These trajectories include motion in line, circle and mixture of both (line and circular). Experimental results describe the successfully implementation of proposed algorithm in prediction/estimation of the Desired Motion Intention (DMI).