Force Calibration of Soft-Sensing Unit for Flexible Exoskeleton

Soft sensors are increasingly used in wearable de-vices as well as soft robotics. This paper introduces a novel, lightweight soft-sensing unit and a new data collection method, aimed at enabling the implementation of soft sensors in robotics. These advancements are designed to replace traditional sen-sors, providing precise data capture and comfortable wearing experience. Through machine learning, soft stretchable sensors originally used for displacement detection have been endowed with the capability to precisely detect tension. The research first involves six participants and two scenarios (manual assistance, motor assistance on treadmill walking), providing high-quality data with different variability for model training. The results demonstrate that the Gated Recurrent Unit (GRU) model out-performs others under comparison in this study, achieving a root mean square error of 2.92 N. Transfer learning is then used to improve the performance of our model under another condition (manual assistance on level ground walking), which achieves 271 % improvement in R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> and maintains consistent performance with data collected from another participant. Our flexible exoskeleton has been tried by a new subject and achieved comfortable assistance without the use of a load cell. Our unit, along with its calibration method presented in this paper, holds great promise for the actual deployment of soft sensors in soft robotics.