Search

Electromyography (EMG) signals have been used in designing muscle-machine interfaces (MuMIs) for various applications, ranging from entertainment (EMG controlled games) to human assistance and human augmentation (EMG controlled prostheses and exoskeletons). For this, classical machine learning methods such as Random Forest (RF) models have been used to decode EMG signals. However, these methods depend on several stages of signal pre-processing and extraction of hand-crafted features so as to obtain the desired output. In this work, we propose EMG based frameworks for the decoding of object motions in the execution of dexterous, in-hand manipulation tasks using raw EMG signals input and two novel deep learning (DL) techniques called Temporal Multi-Channel Transformers and Vision Transformers. The results obtained are compared, in terms of accuracy and speed of decoding the motion, with RF-based models and Convolutional Neural Networks as a benchmark. The models are trained for 11 subjects in a motion-object specific and motion-object generic way, using the 10-fold cross-validation procedure. This study shows that the performance of MuMIs can be improved by employing DL-based models with raw myoelectric activations instead of developing DL or classic machine learning models with hand-crafted features.

With an increasing number of robotic and prosthetic devices, there is a need for intuitive Muscle-Machine Interfaces (MuMIs) that allow the user to have an embodied interaction with the devices they are controlling. Such MuMIs can be developed using machine learning based methods that utilize myoelectric activations from the muscles of the user to decode their intention. However, the choice of the learning method is subjective and depends on the features extracted from the raw Electromyography signals as well as on the intended application. In this work, we compare the performance of five machine learning methods and eight time-domain feature extraction techniques in discriminating between different gestures executed by the user of an EMG based MuMI. From the results, it can be seen that the Willison Amplitude performs consistently better for all the machine learning methods compared in this study, while the Zero Crossings achieves the worst results for the Decision Trees and the Random Forests and the Variance offers the worst performance for all the other learning methods. The Random Forests method is shown to achieve the best results in terms of achieved accuracies (has the lowest variance between subjects). In order to experimentally validate the efficiency of the Random Forest classifier and the Willison Amplitude technique, a series of gestures were decoded in a real-time manner from the myoelectric activations of the operator and they were used to control a robot hand.

Electromyography (EMG) based interfaces are the most common solutions for the control of robotic, orthotic, prosthetic, assistive, and rehabilitation devices, translating myoelectric activations into meaningful actions. Over the last years, a lot of emphasis has been put into the EMG based decoding of human intention, but very few studies have been carried out focusing on the continuous decoding of human motion. In this work, we present a learning scheme for the EMG based decoding of object motions in dexterous, in-hand manipulation tasks. We also study the contribution of different muscles while performing these tasks and the effect of the gender and hand size in the overall decoding accuracy. To do that, we use EMG signals derived from 16 muscle sites (8 on the hand and 8 on the forearm) from 11 different subjects and an optical motion capture system that records the object motion. The object motion decoding is formulated as a regression problem using the Random Forests methodology. Regarding feature selection, we use the following time-domain features: root mean square, waveform length and zero crossings. A 10-fold cross validation procedure is used for model assessment purposes and the feature variable importance values are calculated for each feature. This study shows that subject specific, hand specific, and object specific decoding models offer better decoding accuracy that the generic models.

Adaptive, tendon-driven and affordable prosthetic devices have received an increased interest over the last decades. Prosthetic devices range from body-powered solutions to fully actuated systems. Despite the significant progress in the field, most existing solutions are expensive, heavy, and bulky, or they cannot be used for partial hand amputations. In this paper, we focus on the development of adaptive, tendon-driven, glove-based, affordable prostheses for partial hand amputations and we propose two compact and lightweight devices (a body powered and a motor driven version). The efficiency of the devices is experimentally validated and their performance is evaluated using two different types of tests: i) grasping tests that involve different everyday objects and ii) tests that assess the force exertion capabilities of the proposed prostheses.

Soft, underactuated, and wearable robotic exo-gloves have received an increased interest over the last years. These devices can be used to improve the capabilities of healthy individuals or to assist people that suffer from neurological and musculoskeletal diseases. Despite the significant progress in the field, most existing solutions are still heavy and expensive, they require an external power source to operate, and they are not wearable. In this paper, we focus on the development of an affordable, underactuated, tendon-driven, wearable exo-glove equipped with a novel four-output differential mechanism that provides grasping capabilities enhancement to the user. The device and the differential mechanism are experimentally tested and assessed using three different types of experiments: i) grasping tests that involve different everyday objects, ii) force exertion capability tests that assess the fingertip forces for different types of grasps, and iii) tendon tension tests that estimate the maximum tendon tension that can be obtained by employing the proposed differential. The device considerably improves the grasping capabilities of the user with a weight of 690 g and an operation autonomy of a whole day.

The field of Brain Machine Interfaces (BMI) has attracted an increased interest due to its multiple applications in the health and entertainment domains. A BMI enables a direct interface between the brain and machines and is capable of translating neuronal information into meaningful actions (e.g., Electromyography based control of a prosthetic hand). One of the biggest challenges in developing a surface Electromyography (sEMG) based interface is the selection of the right muscles for the execution of a desired task. In this work, we investigate optimal muscle selections for sEMG based decoding of dexterous in-hand manipulation motions. To do that, we use EMG signals derived from 14 muscle sites of interest (7 on the hand and 7 on the forearm) and an optical motion capture system that records the object motion. The regression problem is formulated using the Random Forests methodology that is based on decision trees. Regarding features selection, we use the following time-domain features: root mean square, waveform length and zero crossings. A 5-fold cross validation procedure is used for model assessment purposes and the importance values are calculated for each feature. This pilot study shows that the muscles of the hand contribute more than the muscles of the forearm to the execution of inhand manipulation tasks and that the myoelectric activations of the hand muscles provide better estimation accuracies for the decoding of manipulation motions. These outcomes suggest that the loss of the hand muscles in certain amputations limits the amputees' ability to perform a dexterous, EMG based control of a prosthesis in manipulation tasks. The results discussed can also be used for improving the efficiency and intuitiveness of EMG based interfaces for healthy subjects.

Classical robotic approaches to tactile object identification often involve rigid mechanical grippers, dense sensor arrays, and exploratory procedures (EPs). Though EPs are a natural method for humans to acquire object information, evidence also exists for meaningful tactile property inference from brief, non-exploratory motions (a 'haptic glance'). In this work, we implement tactile object identification and feature extraction techniques on data acquired during a single, unplanned grasp with a simple, underactuated robot hand equipped with inexpensive barometric pressure sensors. Our methodology utilizes two cooperating schemes based on an advanced machine learning technique (random forests) and parametric methods that estimate object properties. The available data is limited to actuator positions (one per two link finger) and force sensors values (eight per finger). The schemes are able to work both independently and collaboratively, depending on the task scenario. When collaborating, the results of each method contribute to the other, improving the overall result in a synergistic fashion. Unlike prior work, the proposed approach does not require object exploration, re-grasping, grasp-release, or force modulation and works for arbitrary object start positions and orientations. Due to these factors, the technique may be integrated into practical robotic grasping scenarios without adding time or manipulation overheads.