Abstract: Computerized machine knitting offers an attractive fabrication technology for incorporating wearable assistive devices into garments. In this work, we utilized, for the first time, whole-garment knitting techniques to manufacture a seamless fully knitted pneumatic bending actuator, which represents an advancement to existing cut-and-sew manufacturing techniques. Various machine knitting parameters were investigated to create anisotropic actuator structures, which exhibited a range of bending and extension motions when pressurized with air. The functionality of the actuator was demonstrated through integration into an assistive glove for hand grip action. The achieved curvature range when pressurizing the actuators up to 150 kPa was sufficient to grasp objects down to 3 cm in diameter and up to 125 g in weight. This manufacturing technique is rapid and scalable, paving the way for mass-production of customizable soft robotics wearables.

Assoc. Prof. Dr. Ozgur Atalay has been awarded by TUBITAK 2247-A "National Leader Researchers Program".

Omur Fatmanur Erzurumluoglu and Ali Emre Oztas have been awarded by Trainee Researcher Scholarship Program (STAR) initiated by the Scientific and Technological Research Council of Turkey (TÜBİTAK).

Abstract: Real-time human activity recognition is a popular and challenging topic in sensor systems. Inertial measurement units, vision-based systems, and wearable sensor systems are mostly used for gathering motion data. However, each system has drawbacks such as drift error, illumination, occlusion, etc. Therefore, under certain circumstances, they are not efficient alone in activity estimation. To overcome this, hybrid sensor systems were used as an alternative approach in the last decade. In this study, a human activity recognition system is proposed using textile-based capacitive sensors. The aim of the system is to recognize the basic human actions in real-time such as walking, running, squatting, and standing. The sensor system proposed in this study is used to collect human activity data from the participants with different anthropometrics and create an activity recognition system. The performance of the machine learning models is tested on unseen activity data. The obtained results showed the effectiveness of our approach by achieving high accuracy up to 83.1% on selected human activities in real-time.