An identification method of human joint interaction torque based on discrete EMG signals

The interactive joint torque serves as a critical biomechanical parameter for intent recognition in exoskeleton motion control systems, enabling adaptive control capabilities within the human-in-the-loop (HITL) closed-loop framework. While this interactive torque fundamentally differs from the actual output torque of joints, empirical studies have demonstrated a quantifiable linear correlation between these two metrics. Consequently, real-time monitoring of joint output torque provides actionable insights into human motion intention, serving as a critical feedback mechanism for intention-driven control strategies in lower-limb exoskeleton applications.

This paper proposes a method for extracting the interactive joint torque of the human body based on the collection of discrete electromyography (EMG) signals. In order to detect and analyze the interactive joint torque, based on the acquisition of human EMG signals, the human joint motion is discretized within a continuous range using a discrete prediction method. Then, the results of discrete learning are converted into a continuous form to establish a numerical relationship between human muscle movement and interactive joint torque.

This identification method has high accuracy under different motion states of the human body. The mean square error of all experiments is 0.1502, the mean coefficient of determination is 0.8616, and the mean coefficient of correlation is 0.9365.

A discrete prediction technology of human joint interaction torque based on EMG acquisition is established, which is helpful to deeply understand the relationship between muscle activity and joint motion, and provides a feasible method for extracting human joint torque.