Automatic Movement Segmentation of Exoskeleton Data

Abstract :

We present a method to automatically segment manipulation movement demonstrated with an exoskeleton into distinct action units. The automatic segmentation of movement plays an important role in applications such as robotic learning from demonstration [4], as well as in braincomputer interfaces when labels for machine learning methods are needed, e.g. for movement prediction [8]. The presented segmentation method is motivated by the hypothesis that human movement is composed of simple building blocks which can be combined to complex behavior [2]. In manipulation movements, these building blocks are characterized by a bell-shaped velocity profile of the hand [5]. We use this to segment human movement trajectories of manipulation tasks into movement building blocks.

Links:

https://depositonce.tu-berlin.de/handle/11303/8075