The SVEN project produced results in the 1970's and 80's that are still among the best ever seen for simultaneously controlling multiple functions in an arm prosthesis. Their system used "crisp classification" of myoelectric (EMG) signals, and controlled multiple motors in an on/off fashion. Today we see several commercial devices with multiple functions, continually increasing the demand for advanced, coordinated control scehemes.
The present project is a part of a larger project in which we will produc e more advanced techniques through the development of robust methods for simultaneous, proportional multifunction prosthesis control.
Our approach extends "crisp classification" to a scheme where each prosthesis function may be activated to a larger or lesser degree, which may be seen as a continuous mapping from the input space of control signals to an output space of prosthesis movement setpoints. This is non-trivial as the number of input signals and functions increases, because each input signal is correlated to each degree of freedom in a complex manner.
The present sub-project focuses specifically on improving control signal quality. Surface EMG measurements are heavily affected by variations in the contact force between the skin and the electro des, motion and skin impedance. These disturbances cause the prosthesis to exhibit unsolicited behaviour. When moving from on/off control to coordinated, proportional multifunction control, such artifacts may yield the system useless because it requires s ignificantly higher signal and control fidelity than traditional methods. We will therefore construct mathematical models of these artifacts and a physical device that allows the simultaneous measurement of both EMG and artifact-inducing quantities like f orce. By proper, model-based signal processing, we aim at producing high-fidelity control signals that exhibit a significantly better signal-to-noise ratio.