MindArm: Mechanized Intelligent Non-Invasive Neuro-Driven Prosthetic Arm System

Introduction

The provided text discusses the significant need for a low-cost solution to help people with limited arm mobility perform daily activities. Approximately 5.4 million people live with paralysis in the United States, and 57.7 million people globally live with limb amputations. The current technological solutions, such as mind-controlled prosthetic devices or invasive brain surgery, are often very costly and inaccessible for many patients with different financial backgrounds. The text identifies the need for an alternative low-cost solution that can help these patients move their arms and perform desired actions, thereby improving their quality of life.

The provided text discusses the existing solutions and their limitations for addressing the physiological limitations of patients. The solutions can be classified into two categories: non-invasive solutions, such as mind-controlled prosthetic devices, and invasive solutions, such as brain-computer interfaces (BCIs).

Non-invasive solutions, like prosthetic-based solutions using electromyography (EMG) signals, do not require invasive devices inside the patient's body. However, EMG signals may not be present in certain conditions, and other non-invasive solutions like mind-controlled prosthetic devices are typically very costly.

Invasive solutions, which employ devices implanted inside the patient's body, can provide higher-quality information signals. However, these are high-risk procedures, expensive, and often not wireless, making them difficult to maintain.

Given the benefits and weaknesses of the state-of-the-art, the potential solution is to develop a non-invasive EEG-based prosthetic arm with low-cost technology and high accuracy of signal processing. This solution faces several challenges, such as designing an effective signal processing pipeline with low-cost devices, removing environmental noise from EEG signals, and developing an effective algorithm to learn and extract information from the EEG signals.

The paper then introduces the MindArm system, which is a non-invasive, neuro-driven prosthetic arm system that employs a deep neural network (DNN) engine to extract information from EEG signals and translate them into prosthetic arm actions. The paper highlights the novel contributions of the MindArm system, including removing noise in the EEG signals, learning EEG signal features using DNN training, and designing a low-cost prosthetic arm.

Methodology

The paper describes the MindArm system, which uses EEG signals to control a prosthetic arm. The key points are:

EEG Data Collection & Extraction: The system uses the OpenBCI Ganglion board with 4 EEG channels and dry electrodes. This provides a cost-effective way to collect EEG data compared to more expensive commercial devices.

Signal Processing: The EEG signals are processed to extract features in different frequency bands (delta, theta, alpha, beta, gamma). This allows the system to better distinguish between relaxation and concentration states.

Neural Network Training: Several neural network architectures were evaluated, with a Transformer-based model achieving the highest validation accuracy of 97.1%. The training data is dynamically segmented into windows to capture sustained thought patterns.

Prosthetic Design: The prosthetic arm uses servo motors and an elastic mechanism to provide realistic hand and finger movements. A modular design allows for easy maintenance and repairs. A low-cost silicone glove further reduces the overall cost of the prosthetic.

The system aims to provide an affordable and practical mind-controlled prosthetic solution compared to existing commercial offerings.

Results and Discussion

The experimental setup for the prosthetic system includes an array of specialized tools and technologies. Neural signal acquisition is managed using the Ganglion board and GUI from OpenBCI, with data communication facilitated via UDP networking. The PyTorch framework is employed for developing and training the neural network. The system's commands are transmitted through serial communication with the Arduino C++ IDE. The design and prototyping phase utilizes Tinkercad, Fusion360, and custom settings in Prusa Slicer.

The Transformer network's performance is analyzed by interfacing the model output with an Arduino board using serial communication protocols. The Arduino translates the neural network's output into actionable commands for the prosthetic hand. Each predicted action from the network triggers the corresponding movement in the prosthetic hand, demonstrating the system's potential for real-world applications. The deployment showcases the high accuracy of the Transformer network, as reflected in the classification report and confusion matrix, as well as its capability to operate in real time with the physical hardware, offering a seamless transition from prediction to action execution.

The confusion matrix and classification report provide detailed performance metrics for the model, highlighting its ability to differentiate between the prosthetic's actions with a high degree of accuracy on the test dataset. The real-world test cases also showcase good accuracy for the system.

V Conclusion

The paper presents a MindArm methodology to create a low-cost, mind-controlled prosthetic arm solution for people with disabilities. The system uses EEG technology to translate brain signals into intended arm movements, employing a deep neural network model to interpret the brain signals and control the prosthetic arm. The experimental results show the MindArm system achieves positive success rates for three different actions: 90% for idle/stationary, 80% for shaking hands, and 80% for picking up a cup. This demonstrates the MindArm provides a novel, low-cost approach for mind-controlled prosthetic devices.