Deep Heterogeneous Dilation of LSTM for Transient-Phase Gesture Prediction Through High-Density Electromyography: Towards Application in Neurorobotics

Abstract

Deep networks have been recently proposed to estimate motor intention using conventional bipolar surface electromyography (sEMG) signals for myoelectric control of neurorobots. In this regard, Deepnets are generally challenged by long training times (affecting practicality and calibration), complex model architectures (affecting the predictability of the outcomes), and a large number of trainable parameters (increasing the need for big data). Capitalizing on our recent work on homogeneous temporal dilation in a Recurrent Neural Network (RNN) model, this paper proposes, for the first time, heterogeneous temporal dilation in an LSTM model and applies that to high-density surface electromyography (HD-sEMG), allowing for the decoding of dynamic temporal dependencies with tunable temporal foci. In this paper, a 128-channel HD-sEMG signal space is considered due to the potential for enhancing the spatiotemporal resolution of human-robot interfaces. Accordingly, this paper addresses a challenging motor intention decoding problem of neurorobots, namely, transient intention identification. Our approach uses only the dynamic and transient phase of gesture movements when the signals are not stabilized or plateaued, which can significantly enhance the temporal resolution of human-robot interfaces. This would eventually enhance seamless real-time implementations. Additionally, this paper introduces the concept of "dilation foci" to modulate the modeling of temporal variation in transient phases. In this work a high number (e.g., 65) of gestures is included, which adds to the complexity and significance of the understudied problem. Our results show state-of-theart performance for gesture prediction in terms of accuracy, training time, and model convergence.

Deep Augmentation for Electrode Shift Compensation in Transient High-density sEMG: Towards Application in Neurorobotics

“…Regarding the use of machine learning (ML) for peripheral human-machine interfacing, a large volume of literature has been conducted focusing on the processing of surface electromyogram (sEMG) signals for the control of upper limb prostheses and orthoses and through decoding the user’s intended motor task. This field has gone through a rapid progression after the introduction of deep learning models (for example, [3]–[7]). Despite recent algorithmic progress related to deep learning (DL) models as well as mechatronic progress such as soft bionic hands [8] and soft exosuits [9], these technologies mainly work predictably in laboratory settings.…”

“…This has significantly boosted the information rate captured from one muscle and has been (a) processed using computational methods (such as independent component analysis) to enable decomposition of the collected dense signal space into spike train activities deriving motor units in the muscles [20], and (b) more recently as a direct feed into deep neural network models to decode the motor intention of the user with ultimate spatiotemporal resolution [3]. In [3] we have recently proposed a novel deep recurrent neural network, one of the very first deep neural networks on transient-phase high-density EMG data, that can map the dynamic phase of high-density recording, captured using 128 channels from the upper limb (64 flexors and 64 extensors), into a prediction of over 60 classes of gestures. The proposed algorithm was able to make the convergence of the network 20 times faster when compared with conventional networks while providing high accuracy and sensitivity.…”

“…HD-sEMG enables decomposition of the collected dense signal space into spike train activities deriving motor units in the muscles [15]. More recently, HD-sEMG has been used as a direct input into deep neural network models to decode the motor intention of the user with high spatiotemporal resolution [1]. In [1], we have recently proposed a novel deep recurrent neural network, one of the very first deep neural networks on transient-phase high-density EMG data, that can map the dynamic phase of high-density recording captured using 128 channels from the upper limb (64 flexors and 64 extensors), into a prediction of over 60 classes of gestures.…”

“…More recently, HD-sEMG has been used as a direct input into deep neural network models to decode the motor intention of the user with high spatiotemporal resolution [1]. In [1], we have recently proposed a novel deep recurrent neural network, one of the very first deep neural networks on transient-phase high-density EMG data, that can map the dynamic phase of high-density recording captured using 128 channels from the upper limb (64 flexors and 64 extensors), into a prediction of over 60 classes of gestures. Utilizing the new concept of temporal dilation of LSTM, the proposed algorithm was able to make the convergence of the network 20 times faster when compared with conventional networks while providing high accuracy and sensitivity.…”

“…The field has gone through a rapid progression after the introduction of deep learning (DL), for example, [1]- [5]. Despite recent algorithmic progress, a major obstacle to realworld implementation is the need for extensive recalibration and the corresponding sensitivity of the trained model to various experimental conditions.…”