Yikun Gu

An Anthropomorphic Robot Hand Developed Based on Underactuated Mechanism and Controlled by EMG Signals

When developing a humanoid myo-control hand, not only the mechanical structure should be considered to afford a high dexterity, but also the myoelectric (electromyography, EMG) control capability should be taken into account to fully accomplish the actuation tasks. This paper presents a novel humanoid robotic myocontrol hand (AR hand III) which adopted an underactuated mechanism and a forearm myocontrol EMG method. The AR hand III has five fingers and 15 joints, and actuated by three embedded motors. Underactuation can be found within each finger and between the rest three fingers (the middle finger, the ring finger and the little finger) when the hand is grasping objects. For the EMG control, two specific methods are proposed: the three-fingered hand gesture configuration of the AR hand III and a pattern classification method of EMG signals based on a statistical learning algorithm — Support Vector Machine (SVM). Eighteen active hand gestures of a testee are recognized effectively, which can be directly mapped into the motions of AR hand III. An on-line EMG control scheme is established based on two different decision functions: one is for the discrimination between the idle and active modes, the other is for the recognition of the active modes. As a result, the AR hand III can swiftly follow the gesture instructions of the testee with a time delay less than 100 ms.

Switching-State Phase Shift Method for Three-Phase-Current Reconstruction With a Single DC-Link Current Sensor

Instead of directly sampling the phase currents, a technique that adopts a single current sensor in the dc link to reconstruct three-phase currents has arisen. Through this approach, the cost of servo systems can be reduced and the reliability can be improved. However, the duration of an effective vector may be so short that the phase current cannot be measured reliably by the conventional space-vector pulsewidth modulation (PWM) algorithm. In this paper, a new phase-current reconstruction method, switching-state phase shift (SSPS), is proposed based on the PWM pattern modification. The modification is performed by applying phase shifts to the switching-state waveforms of the inverter. The algorithm can realize high-quality reconstruction to the phase current with less impacts on the output current ripples and switching losses. Moreover, it can maximize the linear-modulation region by saturation handling. Measurements on the permanent-magnet synchronous motor servo drives show that the SSPS can offer an attractive performance during both steady-state and dynamic operations in the phase-current reconstruction and regulation.

EMG pattern recognition and grasping force estimation: Improvement to the myocontrol of multi-DOF prosthetic hands

The multi-DOF prosthetic hands myocontrol needs to recognize more hand gestures (or motions) based on myoelectric signals. This paper presents a classification method, which is based on the support vector machine (SVM), to classify 19 different hand gesture modes through electromyographic (EMG) signals acquired from six surface myoelectric electrodes. All hand gestures are based on a 3-DOF configuration, which makes the hand perform like three-fingered. The training performance is very high within each test session, but the cross-session validation is typically low. Acceptable cross-session performance can be achieved by training with more sessions or fewer gesture modes. A fast rhythm muscle contraction is suggested, which can make the training samples more resourceful and improve the prediction accuracy comparing with a relative slow muscle contraction method. For many precise grasp tasks, it is beneficial to the prosthetic hands myocontrol if we can efficiently extract the grasp force directly from EMG signals. Through grasping a JR3 6 dimension force/torque sensor, the force signal applying to the sensor can be recorded synchronously with myoelectric signals. This paper uses three methods, local weighted projection regression (LWPR), artificial neural network (ANN) and SVM, to find the best regression relationship between these two kinds of signals. It reveals that the SVM method is better than ANN and LWPR, especially in the case of cross-session validation. Also, the performance of grasping force estimation based on specific hand gestures is superior to the performance of grasping with random fingers.

Estimation of hand grasp force based on forearm surface EMG

In the force control of multi-functional prosthetic hands, it is important to extract grasp force information besides mode specifications directly from the myoelectric signals. In this paper, a force sensor is adopted to record the hands enveloping force when the hand is performing several grasp modes, synchronously with 6 channels surface electromyography (EMG) which are extracting from the subjects forearm. Three pattern regression methods, locally weighted projection regression (LWPR), artificial neural network (ANN) and support vector machine (SVM) are used to find the best representative relationship of these two kinds of signals. Experimental results show that the SVM method is better than LWPR and ANN, especially in the case of cross-session validation. Also, the force regression performance is better when grasping within several specific modes than grasping randomly. Based on these results, an efficient online prediction of the hand grasp force is present finally, with an accuracy of around 0.9 in squared correlation coefficient (SCC) and 5~10N error over a range of 60N. It can be utilized for the prosthetic hands control to provide a reasonable exerting force reference.

Dynamic training protocol improves the robustness of PR-based myoelectric control

Abstract In pattern recognition (PR)-based myoelectric control schemes, the classifier is generally trained in ideal laboratory conditions, due to which the classification accuracy might be affected by confounding factors such as force variations, limb positions, and inadvertent electromyography (EMG) activation. Many endeavors have been put forward to mitigate this effect by adopting new training protocols that consider only quite a few independent factors. In this note, we propose a dynamic protocol, which embraces multiple EMG variations in data collection, to train a classifier with improved generalization ability. A total of four training protocols are examined, wherein affecting factors like upper-limb movements, contraction levels and inadvertent EMG activations are differently considered. Based on receiver operating characteristic (ROC) analysis, we came up with a new performance metric, ROC area rate (RAR), to directly inspect the accuracy and robustness of the classifiers obtained through different training protocols. Our results show that, compared with the other three protocols, the protocol with dynamic limb postures and dynamic muscle contractions (termed as DPDE) obtains the highest RAR (73.3%, on-way analysis of variance, p 0.005 ). Our results suggest that there is no need to integrate every EMG variation in the training protocol for receiving a robust EMG pattern recognition. Online control experiments with three amputees manipulating a multiple-DOF prosthetic hand also verify our findings.

Dexterous motion recognition for myoelectric control of multifunctional transradial prostheses

Attributed to its superior dexterity, the human’s thumb plays a significant role in the activities of daily living (ADLs). In order to promote the operational capability of the prosthetic hand, state-of-the-art designs tend to integrate multiple degrees of freedom (DOFs) in thumb’s trapeziometacarpal (TM) joint for achieving versatile movements. However, without proper control methods, this fine dexterity of multi-DOF prosthetic hand is hard to be demonstrated. This problem becomes more serious when the control signals, electromyography (EMG), are highly limited on the residual stump of the amputee. Through experiments, the paper examines the feasibility of recognizing 4 thumb motions (flexion/extension and abduction/adduction) together with the other 12 finger- and wrist-related motions under a general surface EMG configuration. After surveying the classification accuracy of several selective motion groups (thumb motions, thumb-involved finger motions, thumb-involved wrist motions, etc.), we report here the current limitations of the prevalent pattern recognition-based EMG control schemes. Considering the severely impaired neuromuscular systems of the amputees, controlling those thumb motions for achieving dexterous operations still faces many challenges. We finally discuss several alternative strategies, including switching, extended physiological proprioception (EPP) and postural synergy, for dealing with this problem. Graphical Abstract

A novel phase current reconstruction method using a single DC-link current sensor

Instantaneous phase currents are required for successful operation of three-phase current controllers. The technique for reconstructing motor phase currents using information from a single current sensor in the dc-link of an inverter has been applied to reduce the cost, weight and volume of system. However, the duration of an active switching state may be so short that the dc-link current cannot be measured reliably. This paper presents an improved PWM modulation strategy named switching state axial translation (SSAT) to overcome this problem. The new SSAT method modifies the PWM voltage vectors by moving the switching state waveforms along the axis when the phase currents cannot be measured. A combination of analytical and experimental results shows that this method offers an attractive phase current reconstruction performance.

Robust EMG pattern recognition in the presence of confounding factors: features, classifiers and adaptive learning

Abstract Traditional electromyogram (EMG) pattern recognition does not take into account the effect of confounding factors, preventing its effective clinical application. In this paper, we investigated EMG pattern recognition of compound confounding factors including electrode shift, force variation, limb posture and temporary drift. We started with feature extraction, to identify suitable representations to resolve classification degradation caused by electrode doffing. Then, we proceeded to examine a variety of classifiers, to achieve a favorable classification accuracy (CA) when switching the control arms (dominant and nondominant) without additional calibration. Lastly, we investigated adaptive learning strategies, to mitigate the decline in CA over a relatively long duration (one day) of use. Two data-collection protocols (electrode doffing and donning, and one-day-long collection) were applied, and EMG datasets collected in these two protocols were arranged into three scenarios for progressive validation of the features, classifiers, and adaptive learning methods. Our results showed that the features extracted from the frequency domain improved the CA by 30% in comparison with their original time-domain representations (14 motions, 67.5 ± 9.8%, p

Adaptive hybrid position force control of dual-arm cooperative manipulators with uncertain dynamics and closed-chain kinematics

Abstract In this paper, a novel adaptive control for dual-arm cooperative manipulators is proposed to accomplish the hybrid position/force tracking in the presence of dynamic and closed-chain kinematic uncertainties. Self-convergent parameter estimation of the grasped objects centre of mass and contact force estimation are incorporated into this systematic scheme. Moreover, internal force and contact force tracking objectives are achieved simultaneously by incorporating into the position tracking formula with proper null-space projection and rotation transformation. Noisy force derivative signals are not required. This adaptive controller is mathematically derived based on Lyapunov stability analysis. Three sets of simulations corresponding to three different situations are presented to verify the effectiveness and superiority of the proposed controller.

Accurate EMG onset detection in pathological, weak and noisy myoelectric signals

Abstract In this paper, we propose an alternative onset detection method dealing with pathological, weak and noisy myoelectric signals. We evaluate our method on simulated, offline EMG signals, which are supposed to be generated from a relatively small number of motor units (MU’s) with various muscle contraction levels and pathological characteristics. These simulated signals were scaled and then superimposed to a standard white noise to obtain various signal conditions (signal noise ratio, SNR). We utilize the Teager-Kaiser Energy (TKE) operator as a fore-processing procedure to highlight amplitude variation on the onset point, and employ two image enhancement technologies, namely, morphological close operator (MCO) and morphological open operator (MOO), as successive post-processing procedures to filter out onset artefacts. A synthesized index for evaluating the method is proposed, which can optimize the parameters according to specific signal conditions. Comparing with other approaches, our method is simple and competitive in accuracy and reliability, especially for the pathological EMG signals in low SNR’s. Result on clinic EMG signals that collected from healthy subjects and patients with amyotrophic lateral sclerosis and myopathy also verifies our design.