Abstract myoelectric control with EMG drive estimated using linear, kurtosis and Bayesian filtering

Three muscle activation estimators: a linear mean-absolute value filter, a recursive Bayesian method, and a kurtosis filter were compared as control approaches for an abstract myoelectric-controlled interface. The linear filter outperformed both the Bayesian and kurtosis methods with respect to participants' overall scores. Despite significantly less efficient trajectories, the Bayesian filter showed a reduction in the time required to reach individual targets. Results demonstrate both that linear methods can outperform more complex filtering techniques, and that real-time kurtosis may be used as an activation estimator.