Skip to main content
Fig. 2 | Journal of NeuroEngineering and Rehabilitation

Fig. 2

From: Learning regularized representations of categorically labelled surface EMG enables simultaneous and proportional myoelectric control

Fig. 2

The effect of contractive regularization on proportionality of control. The effect is illustrated by a simplified example with univariate regressor \({\varvec{e}}=e\) and regressand \({\varvec{y}}=y\). a Plot of calibration data, simulating movement onset, constituted by the sEMG envelope \({e}_{t}^{cal}\) with \(I=1\) channel (upper blue) and the concurrent movement instruction \({y}_{t}^{cal}\) with \(J=1\) DoF (lower red). b Plot of the learned mapping \(\widehat{y}={f}_{1}(e,{\varvec{\theta}})\) performed by an MRL network calibrated by minimizing \({\mathcal{L}}_{i}\) with respect to \({\varvec{\theta}}\). This model approximates a categorical decision threshold and does not enable proportionality. c Plot of the learned mapping \(\widehat{y}={f}_{2}(e,{\varvec{\theta}})\) performed by an MRL network calibrated by instead minimizing \({\mathcal{L}}_{i}+{\mathrm{\alpha }}_{c}{\mathcal{L}}_{c}\) with respect to \({\varvec{\theta}}\). This model produces output which varies smoothly with latent muscle activity and thus enables proportionality

Back to article page