A Multi-sensor Approach for Biomimetic Control of a Robotic Prosthetic Hand

Abstract

Abstract. Robotic prosthetic hands with five digits have become commercially available however their use is limited to a few grip patterns due to the unnatural and unreliable human-machine interface (HMI). The research community has addressed this problem extensively by investigating Pattern Recognition (PR) based surface-electromyography (sEMG) control. This control strategy has been recently commercialized however has yet to show clinical adoption. One of the reasons identified in the literature is due to the sEMG signals that are affected by sweating, electrode shift, ambient noise, fatigue, cross-talk between adjacent muscles, signal drifting, and force level variation. Hence recently the scientific community has started proposing multi-modal sensing techniques as a solution.This study aims to investigate the use of multi-modal sensor approach to control a robotic prosthetic hand by investigating the sparsely studied sensing mechanism called Force Myography (FMG) as a synergist to the conventional technique of sEMG. FMG uses pressure sensors on the surface of a limb to detect the volumetric changes in the underlying musculotendinous complex. This paper presents a custom prosthetic prototype instrumented with sEMG and FMG sensors and tested by a participant with a transradial amputation. Results demonstrate that this multi-sensor approach has the potential to be a valid HMI for prosthesis control.

A survey on the state of the art of force myography technique (FMG): analysis and assessment

“…For further smoothing, some literature tends to use preprocessing steps such as filtering and normalization [ 24 , 37 , 58 ]. Michael et al [ 106 ] used a low-pass filter with a 4 Hz cutoff frequency as well as Carlo et al [ 85 ].…”

“…Using raw FMG signals but with the appropriate machine learning technique may contribute to further increasing the accuracy in predicting and effort the time used for some preprocessing steps. It has been proved by [ 25 ], that raw FMG signals processed with extreme learning machine (ELM) outperformed the raw FMG signals classified using artificial neural network (ANN) [ 65 ], support vector machine (SVM) [ 117 ], and linear discriminant analysis (LDA) [ 5 , 26 , 37 ]. The later methods can only achieve high accuracies with normalized FMG signals.…”

A Review of Force Myography Research and Development

“…While the majority of FMG investigations use participants with intact limbs, a significant portion of these works actually targeted prosthesis control applications. In recent years, researchers successfully tested the FMG approach for predicting the intended actions of participants with trans-radial amputations [14,31,41,42,43,44]. For prosthesis control, the number of intended predicted actions is much less than the one for participants with intact limbs.…”