sEMG-based Static Force Estimation for Human-Robot Interaction using Deep Learning

Abstract

Human-robot interaction (HRI) is a rapidly growing research area and it occurs in many applications including human-robot collaboration, human power augmentation, and rehabilitation robotics. As it is hard to exactly calculate intended motion trajectory, generally, interaction control is applied in HRI instead of pure motion control. To implement the interaction control, force information is necessary and force sensor is widely used in force feedback. However, force sensor has some limitations as 1) it is subject to breakdown, 2) it imposes additional volume and weight to the system, and 3) its applicable places are constrained. In this situation, force estimation can be a good solution. However, if force in static situation should be measured, using position and velocity is not sufficient because they are not influenced by the exerted force anymore. Therefore, we proposed sEMG-based static force estimation using deep learning. sEMG provides a useful information about human-exerting force because it reflects the human intention. Also, to extract the complex relationship between sEMG and force, deep learning approach is used. Experimental results show that when force with maximal value of 63.2 N is exerted, average force estimation error was 3.67 N. Also, the proposed method shows that force onset timing of estimated force is faster than force sensor signal. This result would be advantageous for faster human intention recognition.