Traditional rigid exoskeletons can be challenging to the comfort of wearers and can have large pressure, which can even alter natural hand motion patterns. In this paper, we propose a low-cost soft exoskeleton glove (SExoG) system driven by surface electromyography (sEMG) signals from non-paretic hand for bilateral training. A customization method of geometrical parameters of soft actuators was presented, and their structure was redesigned. Then, the corresponding pressure values of air-pump to generate different angles of actuators were determined to support four hand motions (extension, rest, spherical grip, and fist). A two-step hybrid model combining the neural network and the state exclusion algorithm was proposed to recognize four hand motions via sEMG signals from the healthy limb. Four subjects were recruited to participate in the experiments. The experimental results show that the pressure values for the four hand motions were about −2, 0, 40, and 70 KPa, and the hybrid model can yield a mean accuracy of 98.7% across four hand motions. It can be concluded that the novel SExoG system can mirror the hand motions of non-paretic hand with good performance.

中文翻译：

---

通过表面肌电图进行手部双边训练的软外骨骼手套


传统的刚性外骨骼对佩戴者的舒适度提出了挑战，并且可能会产生很大的压力，甚至会改变自然的手部运动模式。在本文中，我们提出了一种低成本软外骨骼手套（SExoG）系统，由来自非麻痹手的表面肌电图（sEMG）信号驱动，用于双边训练。提出了一种软执行器几何参数定制方法，并重新设计了软执行器的结构。然后，确定气泵产生不同角度的执行器的相应压力值，以支持四种手部动作（伸展、休息、球形握持和握拳）。提出了一种结合神经网络和状态排除算法的两步混合模型，通过健康肢体的 sEMG 信号识别四种手部动作。招募了四名受试者参加实验。实验结果表明，四种手部动作的压力值约为-2、0、40和70 KPa，混合模型可以在四种手部动作中产生98.7%的平均准确度。可以得出结论，新颖的 SExoG 系统能够以良好的性能反映非麻痹手的手部运动。