Development of a Low-Cost Prosthetic Hand Using Electromyography and Machine Learning

arXiv:2411.15533 · cs.RO, cs.HC · Submitted 2024-11-23 · Read on arXiv

Listen

Transcript

Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.

Rosa: Today's paper: "Development of a Low-Cost Prosthetic Hand Using Electromyography and Machine Learning".

Dev: The gist: This project aims to develop a low-cost and effective myoelectric prosthetic hand that would meet the needs of amputees in developing countries.

Rosa: First, who's behind it and why it matters.

Paper summary: Rosa: To recap, this paper is about creating a low-cost prosthetic hand using EMG signals to control it. The main thrust is making it accessible for amputees in developing countries by accurately classifying five different gestures from three muscles >

Dev: They claim they achieved this by using an FFANN architecture, which they found to be computationally low cost for the classification part, and they tested both time and frequency domain features >

Taro: The core claim is that their system can classify these patterns with high accuracy, reaching ninety-four point eight percent in the time domain and ninety-three point two percent in the frequency domain > <ref:2411.15533#pg1>

Rosa: So it’s not just about classification; they designed a physical hand with two degrees of freedom controlled by servo motors and even added a wrist rotation mechanism using two gestures to get that third degree of freedom >

Dev: And they included tactile feedback via force sensors and vibration motors so the user can feel the force being applied to the prosthetic hand >

Taro: The implication is that combining this level of control, including that extra rotational axis and sensory feedback, into a low-cost package is a significant step toward more functional prosthetics for underserved populations >

Rosa: It shows how machine learning can be adapted to create tangible devices that serve very specific needs without needing massive computing power or expensive sensors >

Dev: The authors are focused on the practical application and minimizing cost, which is what makes this particular study relevant for real-world deployment rather than just theoretical research >

Conclusion: Rosa: Thinking about the title, "Development of a Low-Cost Prosthetic Hand Using Electromyography and Machine Learning," it really boils down to using accessible technology to solve a very human problem >

Dev: The authors, Mosab Diab, Ashraf Mohammed, and Yinlai Jiang are focused on proving that this approach works effectively even when resources are limited >

Taro: For someone listening just for the show, what does this actually change about how we think about assistive technology available globally?

Rosa: It means we can start looking at systems that don't require incredibly complex machinery to provide basic functionality, focusing instead on robust signal processing and machine learning models that work on less powerful platforms >

Dev: And it suggests that the combination of classification accuracy and physical degrees of freedom with sensory feedback is achievable within a constrained budget >

Taro: So we’re talking about moving toward devices where the control mechanism is more intuitive, where the user gets immediate feedback, and where that technology can actually be deployed widely >

Department of Electrical and Electronic Engineering, Faculty of Engineering, University of Khartoum · Center for Neuroscience and Biomedical Engineering, The University of Electro-Communications

cs.RO, cs.HC

Submitted: 2024-11-23

Updated: 2026-10-09

Comments: 6 pages, 9 figures

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 71/100

The gist: The gist: This project aims to develop a low-cost and effective myoelectric prosthetic hand that would meet the needs of amputees in developing countries.

Key concepts

Electromyography (EMG): EMG measures the electrical activity generated by muscles when they contract. This paper uses EMG signals recorded from residual limbs to interpret user intent, allowing the prosthetic hand to move based on muscle signals.
Time Domain Features (TDF): TDF involves analyzing raw EMG signals in terms of their time characteristics. The paper found that TDF analysis provided faster computation and a high classification accuracy of 97.25%, making it the preferred method for classifying user gestures.
Frequency Domain Features (FDF): FDF analyzes the frequency content of the EMG signals. This method also performed well, yielding an accuracy of 98.6%. Both TDF and FDF features were compared to determine which was most effective for controlling the robotic hand.

Terminology

Accuracy — 33× in this paper · explained in 1 episode
Feature Extraction — 10× in this paper · explained in 3 episodes
Time Domain Features — 5× in this paper
Frequency Domain Features — 3× in this paper
Linear Discriminant Analysis — 3× in this paper
Artificial Neural Networks — 2× in this paper
Fast Fourier Transform — 2× in this paper
Quantization — 2× in this paper · explained in 13 episodes
Wavelet Transform — 2× in this paper · explained in 2 episodes
Core — 1× in this paper · explained in 2 episodes
EMG Signals Analog to Digital Conversion — 1× in this paper
Mean Absolute Deviation — 1× in this paper
PCA and Uncorrelated Linear Discriminant Analysis — 1× in this paper
Principal Component Analysis — 1× in this paper
Principal Component Analysis (PCA) — 1× in this paper · explained in 5 episodes
Support Vector Machines — 1× in this paper

Summary

The gist: This project aims to develop a low-cost and effective myoelectric prosthetic hand that would meet the needs of amputees in developing countries.

Development Goal and Scope

The primary aim of this project is to develop a low-cost and effective myo-electric prosthetic hand that would meet the needs of amputees in developing countries. The proposed prosthetic hand should be able to accurately classify five different patterns (gestures) using EMG recordings from three muscles and control a robotic hand accordingly. The robotic hand is composed of two servo motors allowing for two degrees of freedom. Furthermore, a wrist rotation mechanism was designed and tested to add significant functionality to the prosthetic. This mechanism is controlled by two of the five gestures, one for each direction. Finally, a tactile sensory feedback system which uses force sensors and vibration motors was developed to enable sensation of the force inflicted on the hand for the user.

Signal Acquisition and Pre-processing

The signal acquisition process involves using surface electrodes to conduct small currents from the muscle to pre-processing units. These electrodes can be grouped into wet and dry electrodes. Dry Electrodes: Can be used multiple times yielding same accuracy and are used in long term applications such as prosthetics. Raw EMG signals range between +/- 5000 microvolts and their frequency contents range between 6 and 500 Hz. Two techniques must be applied to the raw signals, which are amplification using instrumentation amplifiers and noise cancellation using special circuitry and filters.

Feature Extraction Methods

The choice of feature space is dictated by how well it can cluster and categorize input data. Feature spaces can be mainly categorized into Time Domain Features (TDF) and Frequency Domain Features (FDF).

Classification System Architecture

The classification system utilizes a Feedforward Artificial Neural Network (FFANN) architecture. This computational architecture consists of an input layer, hidden layer/s, and an output layer. FFANN are considered computationally low cost when classifying the data. For frequency domain classification, a network with a 24-neuron input layer, a 10-neuron hidden layer with tansig transfer function, and a 5-neuron output layer with softmax transfer function was used. For time domain classification, the same neural network structure was used with nine neurons as input layer.

Prosthetic Hand Mechanism and Feedback

The prosthetic hand itself is 3D printed using polylactic acid and contains two servo motors. One motor is connected to the thumb and the other to the four remaining fingers allowing for two degrees of freedom. To add functionality, a mechanism that allows for user-controlled wrist rotation was designed using servo motor steel ball bearings and custom-made 3D printed parts. This added a third degree of freedom to the overall design. Tactile sensory feedback is provided by force sensing resistors (FSRs) on each fingertip and coin vibration motors. Three motors were used with each motor operated between 0-2.5V with 2.5V being maximum vibration.

Performance Comparison and Conclusion

The results showed an average classification accuracy of 97.25% for the time domain analysis and 98.6% for the frequency domain analysis. The authors in [16] reported an average classification accuracy reached over three datasets for six gestures was 96.38. It is clearly visible that the methods provided in this paper yield better results with fewer resources than what is current in the field. The total cost of the components of this prosthetic hand was 259.86 USD<ref:2411.

Improvements for AI systems

Improve classification robustness by implementing a hybrid feature extraction strategy that combines aspects of both domains, as The choice of feature space is dictated by how well it can cluster and categorize input data [2]. This would allow the system to leverage both Time Domain Features (TDF) and Frequency Domain Features (FDF) simultaneously, potentially mitigating weaknesses in either domain when faced with noisy or complex muscle signals.
Implement a computationally lighter neural network architecture by exploring transfer functions beyond the current tansig and softmax functions, as the paper notes that FFANNs are flexible and results change drastically by modifying these components [10]. This optimization could lead to a faster response time on the Arduino, aligning with the goal of maintaining faster computation and response for the time domain analysis.
Enhance user acceptance by integrating a more sophisticated sensory feedback mechanism based on advanced tactile sensation modeling. Instead of just using FSRs with vibration motors, explore methods to create more nuanced feedback that better simulates the force inflicted on the hand for the user, potentially moving beyond simple threshold-based operation described in Section IV.C.

Abstract

Electromyography (EMG) is a measure of muscular electrical activity and is used in many clinical/biomedical disciplines and modern human computer interaction. Myo-electric prosthetics analyze and classify the electrical signals recorded from the residual limb. The classified output is then used to control the position of motors in a robotic hand and a movement is produced. The aim of this project is to develop a low-cost and effective myo-electric prosthetic hand that would meet the needs of amputees in developing countries. The proposed prosthetic hand should be able to accurately classify five different patterns (gestures) using EMG recordings from three muscles and control a robotic hand accordingly. The robotic hand is composed of two servo motors allowing for two degrees of freedom. After establishing an efficient signal acquisition and amplification system, EMG signals were thoroughly analyzed in the frequency and time domains. Features were extracted from both domains and a shallow neural network was trained on the two sets of data. Results yielded an average classification accuracy of 94.8% and 93.2% for the time and frequency domains respectively. Furthermore, results showed a faster computation and response for the time domain analysis; hence, it was adopted for the classification system. A wrist rotation mechanism was designed and tested to add significant functionality to the prosthetic. The mechanism is controlled by two of the five gestures, one for each direction. Which added a third degree of freedom to the overall design. Finally, a tactile sensory feedback system which uses force sensors and vibration motors was developed to enable sensation of the force inflicted on the hand for the user.