Enhancing Back-Support Exoskeleton Versatility based on Human Activity Recognition

Designing and testing exoskeletons specifically for one task does not capture the complexity of real environments where exoskeleton versatility is very important. In contrast to passive exoskeletons, quasi-passive and active ones allow to switch off assistance if the actuators are hindering user's movements. However, the assistance switching relies on user manual input and this affects working performances. In this work we are interested in enhancing back-support exoskeletons versatility and to introduce an automatic switching strategy. We propose to use Support Vector Machines for Human Activity Recognition, namely walking, bending and standing. Training and testing of the classifier are based on experimental data collected from ten healthy subjects divided into two sessions with different protocols, in order to evaluate the best one. Results show that the approach is promising with a high level of accuracy (~ 94%), precision (e.g, ~ 94% for bending) and recall (e.g, ~ 91% for walking).