Role of Operator Muscle Coactivation towards Intuitive Interaction with Haptic Assist Devices

Abstract

The current paper addresses a central question at the basis of our research into the control of haptic assist devices: is kinetic sensor information sufficient for the intuitive control of haptic assist devices during physical human-robot interaction (pHRI) Our study proposes the addition of operator physiological data in the form of muscle cocontraction (or coactivation) information, obtained from processing surface electromyographic (sEMG) sensor data from the operator’s arm muscles. Muscle coactivation has shown in previous research to correlate to the operator’s endpoint stiffness, an important system variable not directly observable. Results from target tracking experiments conducted on 10 subjects using a one-degree-of-freedom (1-DoF) haptic device show a muscle cocontraction variable as the best feature to predict the motor intent of the human operator. In addition, using muscle cocontraction data in tandem with kinetic data to train a Support Vector Machine (SVM) classifier offline results in an average cross-validation accuracy of 99%, compared to an accuracy of 89% with kinetic data only. Furthermore, statistically significant trends in cocontraction levels were observed in different stability conditions (trajectory tracking experiments, 10 subjects, p < 0.001) and across operator motor intent (target following experiments, 10 subjects, p < 0.001). These results strongly underline the importance of adding operator physiological information to the sensor data set in order to enhance the control of haptic assist devices.