BRAVE: Brain-Controlled Prosthetic Arm with Voice Integration and Embodied Learning for Enhanced Mobility

TL;DR

BRAVE addresses non-invasive EEG-based prosthetic control for upper-limb amputees by integrating ensemble deep learning (LSTM, CNN) and a RF classifier with robust ICA+CSP artifact rejection, ASR-driven voice switching, and a HITL correction loop. The system demonstrates real-time control with a latency of about 150 ms and achieves roughly 96% classification accuracy across multiple subjects on embedded-capable hardware, using an OpenBCI-based EEG pipeline and a 3D-printed, 3-DOF prosthetic arm. By combining EEG-driven motor intent with intuitive voice commands for mode switching and a human-in-the-loop, BRAVE enhances robustness, usability, and adaptability in real-world settings. The approach has potential to improve accessibility and effectiveness of neuroprosthetics, particularly in resource-constrained environments where cost and portability are critical.

Abstract

Non-invasive brain-computer interfaces (BCIs) have the potential to enable intuitive control of prosthetic limbs for individuals with upper limb amputations. However, existing EEG-based control systems face challenges related to signal noise, classification accuracy, and real-time adaptability. In this work, we present BRAVE, a hybrid EEG and voice-controlled prosthetic system that integrates ensemble learning-based EEG classification with a human-in-the-loop (HITL) correction framework for enhanced responsiveness. Unlike traditional electromyography (EMG)-based prosthetic control, BRAVE aims to interpret EEG-driven motor intent, enabling movement control without reliance on residual muscle activity. To improve classification robustness, BRAVE combines LSTM, CNN, and Random Forest models in an ensemble framework, achieving a classification accuracy of 96% across test subjects. EEG signals are preprocessed using a bandpass filter (0.5-45 Hz), Independent Component Analysis (ICA) for artifact removal, and Common Spatial Pattern (CSP) feature extraction to minimize contamination from electromyographic (EMG) and electrooculographic (EOG) signals. Additionally, BRAVE incorporates automatic speech recognition (ASR) to facilitate intuitive mode switching between different degrees of freedom (DOF) in the prosthetic arm. The system operates in real time, with a response latency of 150 ms, leveraging Lab Streaming Layer (LSL) networking for synchronized data acquisition. The system is evaluated on an in-house fabricated prosthetic arm and on multiple participants highlighting the generalizability across users. The system is optimized for low-power embedded deployment, ensuring practical real-world application beyond high-performance computing environments. Our results indicate that BRAVE offers a promising step towards robust, real-time, non-invasive prosthetic control.

Figures (12)

• Figure 1: BRAVE System Overview: The methodology consists of three stages: (A) EEG Data Collection & Preprocessing, where EEG signals are captured using the OpenBCI Mark IV headset, then amplified, filtered, and streamed for analysis and dataset generation; (B) Neural Network Training, where EEG data is classified using an ensemble of DL models, with the window size optimized and models stored in the cloud for deployment; and (C) Prosthetic Arm Design & Control, where the models are deployed on hardware to classify live EEG data, generating control commands for prosthetic movement. The 3D-printed prosthetic arm, tailored to user needs, operates with 3 DOF via an Arduino-controlled servo system.
• Figure 2: EEG Data Collection Process: (A) The 10-20 system for EEG electrode placement, highlighting standard locations for brainwave detection. (B) The Ultracortex Mark IV headset equipped with 16 channels for comprehensive EEG acquisition.
• Figure 3: (a) Frequency responses of the 6th order Butterworth bandpass filter (0.5–45Hz) and the 50Hz notch filter, showing their attenuation profiles. (b) Power Spectral Density (PSD) of the original and filtered signals, highlighting the reduction of noise and elimination of the 50Hz interference after filtering. (c) Time-domain plot of the original EEG signal with alpha and beta waves, noise, and 50Hz interference. (d) Time-domain plot of the filtered EEG signal after applying both filters, demonstrating improved signal clarity and preservation of key EEG components.
• Figure 4: Using Lab Streaming Layer (LSL) Networking Kothe2024 instead of UDP significantly enhances system performance for EEG-based prosthetic control. LSL facilitates real-time streaming and synchronization of EEG data across devices, providing a consistent sample rate around 125 Hz—crucial for capturing rapid EEG dynamics necessary for accurate neural pattern detection and precise prosthetic control.
• Figure 5: ICA Component Topographies (IC 000 – IC 011): Scalp maps of the 12 independent components extracted from the 16-channel, 1–40 Hz-filtered EEG. Red (positive weights) and blue (negative weights) indicate each component’s contribution across electrodes; darker colours represent larger absolute loadings. Components with strong, focal frontal activity (e.g. IC 001) are characteristic of ocular artefacts, whereas more distributed bilateral patterns (e.g. IC 004, IC 007) are likely of neural origin. These maps guide manual rejection by highlighting artefact-dominated ICs.