Use of Deep Learning for Position Estimation and Control of Soft Glove

Abstract

Soft wearable robots comprised of deformable materials have recently attracted much attention in the field of applications for its lightness and elasticity. However, there exist some limitation to the control of wearable robots due to the complexity of the model and the conditions of the wearer. In this paper, we propose a learning-based position control method of soft wearable glove using a deep neural network (DNN). To analyze our proposed method, we fabricated a soft pneumatic glove and a control board for the glove based on open hardware platform data. With our developed system, we collected the pressure and position data of the soft glove using a Leap Motion sensor to train our soft glove position network (SGPN). Along with our proposed DNN model, we could enable open-loop control on the joint positions of the soft glove by supplying pressure to the actuator without prior knowledge of the wearer or the wearable robot such as hand size of the wearer or the stiffness of pneumatic actuator.