Identification of the best strategy to command variable stiffness using electromyographic signals

Objective. In the last decades, many EMG-controlled robotic devices were developed. Since stiffness control may be required to perform skillful interactions, different groups developed devices whose stiffness is real-time controlled based on EMG signal samples collected from the operator. However, this control strategy may be fatiguing. In this study, we proposed and experimentally validated a novel stiffness control strategy, based on the average muscle co-contraction estimated from EMG samples collected in the previous 1 or 2 s. Approach. Nine subjects performed a tracking task with their right wrist in five different sessions. In four sessions a haptic device (Hi-5) applied a sinusoidal perturbing torque. In Baseline session, co-contraction reduced the effect of the perturbation only by stiffening the wrist. In contrast, during aided sessions the perturbation amplitude was also reduced (mimicking the effect of additional stiffening provided by EMG-driven robotic device) either proportionally to the co-contraction exerted by the subject sample-by-sample (Proportional), or according to the average co-contraction exerted in the previous 1 s (Integral 1s), or 2 s (Integral 2s). Task error, metabolic cost during the tracking task, perceived fatigue, and the median EMG frequency calculated during a sub-maximal isometric torque generation tasks that alternated with the tracking were compared across sessions. Main results. Positive effects of the reduction of the perturbation provided by co-contraction estimation was identified in all the investigated variables. Integral 1s session showed lower metabolic cost with respect to the Proportional session, and lower perceived fatigue with respect to both the Proportional and the Integral 2s sessions. Significance. This study's results showed that controlling the stiffness of an EMG-driven robotic device proportionally to the operator's co-contraction, averaged in the previous 1 s, represents the best control strategy because it required less metabolic cost and led to a lower perceived fatigue.