Generation of real-time control commands from EEG signals
Brain-computer interfaces (BCI) have been studied for decades for their potential to control devices by monitoring brain activity. Specifically, MI-based BCI systems (Motor Imagery) have proven highly effective in the field of neurorehabilitation, as the brain patterns generated closely resemble tho...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis de maestría |
| Fecha de publicación: | 2025 |
| País: | España |
| Institución: | Universitat Politècnica de Catalunya (UPC) |
| Repositorio: | UPCommons. Portal del coneixement obert de la UPC |
| Idioma: | inglés |
| OAI Identifier: | oai:upcommons.upc.edu:2117/424921 |
| Acceso en línea: | https://hdl.handle.net/2117/424921 |
| Access Level: | acceso abierto |
| Palabra clave: | Brain-computer interfaces Electroencephalography Interfícies cervell-ordinador Electroencefalografia Àrees temàtiques de la UPC::Enginyeria biomèdica |
| Sumario: | Brain-computer interfaces (BCI) have been studied for decades for their potential to control devices by monitoring brain activity. Specifically, MI-based BCI systems (Motor Imagery) have proven highly effective in the field of neurorehabilitation, as the brain patterns generated closely resemble those produced during actual movements. This work proposes a protocol designed to acquire and analyze BCI data using the Bitbrain Hero Helmet device, aiming to explore its use in practical applications. The project is part of POSMOFYA, an acronym for the Hybrid Platform of OrthosisWheelchair to ensure compatibility of Mobility, Functionality, and Acceptability for use in domestic environments. This hybrid platform, as its name suggests, integrates an automated wheelchair and a robotic arm. The ultimate goal is to enable efficient and practical control in domestic environments. The developed protocol is based on principles of robustness and precision, aiming to enhance the ability to recognize users’ movement intentions. This work is structured into several phases, beginning with the experimental design. Using the PsychoPy software and the LabStreaming Layer (LSL) protocol, motor imagery tasks were defined focusing on two specific actions: right wrist flexion and left wrist flexion. These tasks were designed to generate brain patterns associated with clear movement intentions, with the goal of training a Machine Learning (ML) system. The strategy ensures precise and consistent capture of brain signals, essential for advancing the development of practical applications within the project framework. The results reflect the challenges of working with EEG signals. Although the Decision Tree algorithm applied to Common Spatial Patterns (CSP) achieved a validation accuracy of 42.9% in offline mode, and K-Nearest Neighbors (KNN) reached 51.0% accuracy with literature-based features, the predictions were biased towards a single class. This highlights limitations in signal discrimination and variability between sessions. Despite the difficulties, the project has provided valuable lessons: the need to improve EEG systems and explore advanced techniques such as Regularized CSP or hybrid methods with deep learning to increase robustness and precision. Additionally, the basic steps of the online pipeline were validated, indicating that better optimization could result in more reliable practical applications. These findings not only highlight the field’s challenges but also point to opportunities for improvement in future research. |
|---|