Continuous Estimation of Human Multi-Joint Angles From sEMG Using a State-Space Model

Abstract

-Due to the couplings among joint-relative muscles, it is a challenge to accurately estimate continuous multi-joint movements from multi-channel sEMG signals. Traditional approaches always build a nonlinear regression model, such as artificial neural network, to predict the multijoint movement variables using sEMG as inputs. However, the redundant sEMG-data are always not distinguished; the prediction errors cannot be evaluated and corrected online as well. In this work, a correlation-based redundancysegmentation method is proposed to segment the sEMG-vector including redundancy into irredundant and redundant subvectors. Then, a general state-space framework is developed to build the motion model by regarding the irredundant subvector as input and the redundant one as measurement output. With the built statespace motion model, a closed-loop prediction-correction algorithm, i.e., the unscented Kalman filter (UKF), can be employed to estimate the multi-joint angles from sEMG, where the redundant sEMG-data are used to reject model uncertainties. After having fully employed the redundancy, the proposed method can provide accurate and smooth estimation results. Comprehensive experiments are conducted on the multi-joint movements of the upper limb. The maximum RMSE of the estimations obtained by the proposed method is 0.16±0.03, which is significantly less than 0.25±0.06 and 0.27±0.07 (p < 0.05) obtained by common neural networks.

A semantics-based approach to sensor data segmentation in real-time Activity Recognition

“…In the data-driven (DD) [7,8] approach, activity models are generated after processing pre-recorded datasets using generative or discriminative classification techniques. The popular generative modelling methods are Bayesian networks, partial Markov decision process (POMDP) [9], a variation of Markov model to model action sequences as finite states with their transitional probabilities and continuous state-space model (CSSM) [10]. Whereas, conditional random field (CRF) and support vector machine (SVM) are widely used as discriminative methods to improve the accuracy and performance of the activity recognition [11].…”

Gaussian Process-Integrated State Space Model for Continuous Joint Angle Prediction from EMG and Interactive Force in a Human-Exoskeleton System

“…The system performance, however, is largely limited by the output of human intent learning and prediction [3]. In many human-robot systems, continuous human motion intents instead of a limited number of classified motion patterns have to be identified and further reconstructed as control commands to the robot so that the robot could match human motion intents and perform efficient assistances [4]. Thus, how to determine the human motion intent in terms of continuous joint angle has become the key issue.…”

“…When the history of the movements reflects the inherent dynamics of human motion [20], joint angle could also be predicted using autoregressive models. To combine the advantages of both and achieve continuous, intuitive and naturalistic understanding of human motion intent, nonlinear autoregressive with exogenous inputs (NARX) would be a better choice and has begun to be applied in the field of joint angle prediction [4,21,22]. Its application combines the non-linear spatio-temporal correlation structure of natural human movements with muscle-driven control signals to exploit the best of both worlds [23].…”

“…As the quality of sEMG signals is very sensitive to muscle fatigue, sweat, and changing electrode characteristics, a proper processing method is needed and the application of on-line learning approaches, such as the GP and ANN as just introduced, would be preferred. In addition, sEMG signals are also used as the measurement output in SSMs for human joint angle prediction, such as the redundant subsectors of the sEMG-vector [4] and other extracted EMG features, including waveform length (WL) and zero crossing (ZC) [27]. Meanwhile, interactive force, which is a direct end-effector reflection in human-exoskeleton systems, can be regarded as an important indicator of transparency between a human and robot [28,29].…”

“…In practical applications, the GP-based NARX model and back propagation neural network (BPNN) are applied to provide better flexibility and adaptivity. Compared with the existing SSMs for continuous joint angle prediction [4,27], our main contributions lie in three aspects. First, interactive force is fused with the EMG signal in the SSM instead of using extracted EMG signal features alone.…”