Automatic support control of an upper body exoskeleton — Method and validation using the Stuttgart Exo-Jacket. (4th September 2020)

• Record Type: Journal Article
• Authors: Singer, Raphael; Maufroy, Christophe; Schneider, Urs
• Is Part Of: Wearable technologies. Volume 1(2021)
• Publication Date: 2020-09-04
• Subjects: Exoskeletons -- Human-Robot Interaction -- Physical Human-Robot Interactive Controllers -- Industry -- Control; Wearable technology -- Periodicals; Robotics in medicine -- Periodicals; 610.28
• DOI: 10.1017/wtc.2020.1
• Languages: English
• ISSNs: 2631-7176

Abstract

Abstract: Although passive occupational exoskeletons alleviate worker physical stresses in demanding postures (e.g., overhead work), they are unsuitable in many other applications because of their lack of flexibility. Active exoskeletons that are able to dynamically adjust the delivered support are required. However, the automatic control of support provided by the exoskeleton is still a largely unsolved challenge in many applications, especially for upper limb occupational exoskeletons, where no practical and reliable approach exists. For this type of exoskeletons, a novel support control approach for lifting and carrying activities is presented here. As an initial step towards a full-fledged automatic support control (ASC), the present article focusses on the functionality of estimating the onset of user's demand for support. In this way, intuitive behavior should be made possible. The combination of movement and muscle activation signals of the upper limbs is expected to enable high reliability, cost efficiency, and compatibility for use in industrial applications. The functionality consists of two parts: a preprocessing—the motion interpretation—and the support detection itself. Both parts were trained with different subjects, who had to move objects. The functionality was validated both in the cases of (A) an unknown subject performing known tasks and (B) a known subject performing unknown tasks. The functionality showed sound results as it achieved a high accuracy ( $$ 95\% $$ ) in training. In addition, the first validation results showed that this functionality is useful for integration in an appropriately adapted ASC and can then enable comfortable working.