Real-time Continuous Hand Motion Myoelectric Decoding by Automated Data Labeling

Abstract

In this paper an automated data labeling (ADL) neural network was proposed to streamline dataset collecting for real-time predicting the continuous motion of hand and wrist, these gestures are only decoded from a surface electromyography (sEMG) array of eight channels. Unlike collecting both the bio-signals and hand motion signals as samples and labels in supervised learning, this algorithm only collects the unlabeled sEMG into an unsupervised neural network, in which the hand motion labels are auto-generated. The coefficient of determination (r2) for three DOFs, i.e. wrist flex/extension, wrist pro/supination, hand open/close, was 0.86, 0.89 and 0.87 respectively. The comparison between real motion labels and auto-generated labels shows that the latter has earlier response than former. The results of Fitts' law test indicate that ADL has capability of controlling multi-DOFs simultaneously even though the training set only contains sEMG data from single DOF gesture. Moreover, no more hand motion measurement needed which greatly helps upper-limb amputee imagine the gesture of residual limb to control a dexterous prosthesis.

Real-Time Wrist Motion Decoding With High Framerate Electrical Impedance Tomography (EIT)

“…Firstly, the TPs of the sequential control scheme based on EIT (1.0269±0.0971 bits/s and 1.0095±0.0931 bits/s without and with sensor re-donning, respectively) outperformed that based on sEMG (0.66±0.06 bits/s in [11]). The TP of the simultaneous control scheme based on EIT (0.9323±0.0869 bits/s) was comparable to those based on sEMG (1.19±0.08 bits/s in [11], 0.87±0.18 bits/s in [12]). The protocols and decoded motions of the studies [11], [12] were similar to ours.…”

“…I and there were eight targets in different orientational directions for each ID. The difficulty levels and setups of the GUI were similar to that of previous studies on muscle-signal-based interfaces [11], [12]. In this experiment, the subjects wore the sensing front-end with the electrodes aligned at the most prominent part of the left forearms.…”

“…The stimulation and measurement mode of our hardware system with parallel sampling. The number means the electrode serial number (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15)(16). The I in red color shows the current injection (pattern).…”

“…Besides investigating the accuracies and time latency separately, researchers in this field used Fitts' law test to evaluate upper-limb motion decoding performances [10]- [13], which evaluated the speed-accuracy tradeoff of multi-DoF kinematics decoding with a throughput (TP) score. For instance, the averaged TP scores ranged from 0.66±0.06 bits/s to 1.19±0.08 bits/s, by decoding the motions of wrist flexion/extension and radial/ulnar deviation based on sEMG [11], [12]. Another critical issue is to produce high performance meanwhile robust and adaptive to external disturbances, which is a research trend of the field in recent few years [14], [15].…”

Feasibility of Using Wearable EMG Armbands combined with Unsupervised Transfer Learning for Seamless Myoelectric Control

“…The proposed approach, further tailored for specific tasks or applications will confer more robust performance of unsupervised transfer learning. Furthermore, the framework employed in this study can be readily extended to more advanced myoelectric control applications, such as for more complex or dynamic human behavior, e.g., especially when integrated with combined EMG and inertial measurement unit data [41], as well as online or real-time control requiring minimal delay [42][43][44][45].…”