Assessment of an On-board Classifier for Activity Recognition on an Active Back-Support Exoskeleton

Abstract

Despite the growing interest, the adoption of industrial exoskeletons may still be held back by technical limitations. To enhance versatility and promote adoption, one aspect of interest could be represented by the potential of active and quasi-passive devices to automatically distinguish different activities and adjust their assistive profiles accordingly. This contribution focuses on an active back-support exoskeleton and extends previous work proposing the use of a Support Vector Machine to classify walking, bending and standing. Thanks to the introduction of a new feature-forearm muscle activity-this study shows that it is possible to perform reliable online classification. As a consequence, the authors introduce a new hierarchically-structured controller for the exoskeleton under analysis.