[Paper Review] Functional Variable Selection for EMG-based Control of a Robotic Hand Prosthetic

Nazmul Islam, Jonathan D. Stallings+4 more|arXiv (Cornell University)|May 8, 2018

EEG and Brain-Computer Interfaces2 references3 citations

TL;DR

This paper proposes a functional varying coefficient model that uses historical EMG signals as functional predictors to predict finger and wrist velocities in robotic hand prosthetics, incorporating biomechanical constraints through a bivariate coefficient function that varies with current limb position. The key contribution is the SAFE-gLASSO method, a two-step functional variable selection procedure that achieves high predictive accuracy with minimal false positives, enabling real-time, interpretable control of prosthetic limbs.

ABSTRACT

State-of-the-art robotic hand prosthetics generate finger and wrist movement through pattern recognition (PR) algorithms using features of forearm electromyogram (EMG) signals, but re- quires extensive training and is prone to poor predictions for conditions outside the training data (Peerdeman et al., 2011; Scheme et al., 2010). We propose a novel approach to develop a dynamic robotic limb by utilizing the recent history of EMG signals in a model that accounts for physiological features of hand movement which are ignored by PR algorithms. We do this by viewing EMG signals as functional covariates and develop a functional linear model that quantifies the effect of the EMG signals on finger/wrist velocity through a bivariate coefficient function that is allowed to vary with current finger/wrist position. The model is made par- simonious and interpretable through a two-step variable selection procedure, called Sequential Adaptive Functional Empirical group LASSO (SAFE-gLASSO). Numerical studies show excel- lent selection and prediction properties of SAFE-gLASSO compared to popular alternatives. For our motivating dataset, the method correctly identifies the few EMG signals that are known to be important for an able-bodied subject with negligible false positives and the model can be directly implemented in a robotic prosthetic.

Motivation & Objective

• To develop a more intuitive and accurate control system for robotic hand prosthetics by leveraging the temporal dynamics of forearm electromyogram (EMG) signals.
• To overcome limitations of existing pattern recognition methods that ignore physiological constraints, suffer from overfitting, and reduce EMG data to summary features.
• To create a parsimonious, interpretable model that accounts for the varying influence of EMG signals on movement based on current limb position.
• To develop a novel variable selection method, SAFE-gLASSO, that identifies the most relevant EMG signals with low false positive rates.
• To enable direct, real-time implementation of the model in robotic prosthetic systems with minimal post-processing.

Proposed method

• Model EMG signals as functional covariates and use a varying coefficient functional linear model where the coefficient function depends on the current finger/wrist position.
• Represent the bivariate coefficient function using a tensor product of rich basis expansions to capture complex, non-linear relationships.
• Apply a combined multi-dimensional smoothing and sparsity penalty to estimate the functional coefficients while enforcing model parsimony.
• Develop a two-step sequential adaptive functional empirical group LASSO (SAFE-gLASSO) procedure for variable selection, prioritizing groups of EMG signals based on their predictive contribution.
• Use data splitting to assess predictive performance and validate model generalization.
• Employ cross-validation to select tuning parameters for the penalized estimation, ensuring optimal balance between fit and sparsity.

Experimental results

Research questions

• RQ1 Can a functional varying coefficient model that incorporates biomechanical constraints improve prediction accuracy in EMG-based robotic prosthetic control?
• RQ2 How can functional variable selection be effectively performed when EMG signals are continuous, high-dimensional curves rather than scalar predictors?
• RQ3 Does the SAFE-gLASSO method outperform standard group LASSO and other variable selection techniques in identifying truly relevant EMG signals with minimal false positives?
• RQ4 To what extent does the model’s predictive performance remain robust when the functional predictors are observed with measurement error or on sparse grids?
• RQ5 Can the selected EMG signals and estimated functional coefficients be directly implemented in real-time prosthetic control systems?

Key findings

• SAFE-gLASSO successfully identified the few EMG signals known to be critical for finger and wrist movement in an able-bodied subject with negligible false positives.
• The estimated functional coefficient functions were relatively sparse and interpretable, reflecting known physiological relationships between muscle activity and joint movement.
• The model demonstrated excellent predictive performance, significantly outperforming standard selection methods in numerical studies.
• The method maintained strong variable selection accuracy even when functional predictors were estimated from noisy or sparse data, indicating robustness to measurement error.
• The final model required minimal data processing for real-time prediction, making it suitable for direct implementation in robotic prosthetic systems.
• The approach is extendable to multivariate functional coefficients and can be adapted to account for posture-dependent variations in EMG effects.