Force estimation for human–robot interaction using electromyogram signals from varied arm postures

Abstract

Abstract In this paper, a system for force estimation based on surface electromyography signals measured from the eight channels of the Myo armband is presented. We evaluated nineteen regression models to continuously estimate force in three scenario cases to cover the natural movement in two degrees of freedom planer rehabilitation mobile robots. The best estimation model that could overcome the challenge in a variety of different scenarios was determined. Based on the experimental results, the Gaussian process regression model performed best, giving a root mean square error in the overall range of 1.18–1.77 N. Additionally, the results showed that the exponential algorithm outperformed other solutions, significantly reducing the force estimation error.

Keywords

• sEMG
• Force estimation
• Regression models
• Human–robot interaction