Localized Extreme Learning Machine for online inverse dynamic model estimation in soft wearable exoskeleton

In recent years, actuation technology have been increasingly developed new fields and utilized widely in applications differing from automation and industry , but also robotic rehabilitation, haptics and wearable exoskeleton devices where safety, limitation of peak forces and gentle interaction are extremely important. To date, several examples of robotic applications have been designed to address the demanding needs of these disciplines that require the compliance in actuation and manipulation. However, the control performance is still limited due to lack of accuracy in robotic dynamics model and unmodeled nonlinearities such as friction. In such cases, estimating inverse dynamic model from collected data will provide an interesting alternative solution in order to achieve the compliance interaction and the good performance in position tracking. In this paper, an algorithm for online robotic inverse dynamics learning is proposed and explained using localization approach combined with Extreme Learning Machine.