Leveraging Edge Impulse for Multi-Class Error-Related Potentials: Edge AI Brain-Robot Interface for Correcting Navigation Errors in Assistive Mobile Robot Wheelchairs

This study explores the integration of BrainComputer Interface (BCI) technology with microcontroller chips to be used for real-time signal processing and event classification, aiming to create a standalone and efficient system deployed as machine learning models on edge devices, optimizing classification pipelines to process Error-Related Potential (ErrP)-based EEG signals. All classification models are multi-class classifiers to differentiate among four classes: 'no navigation error' and three levels of navigation error severity (small, medium, and large). The main aim is to investigate the feasibility of utilizing edge computing for real-time signal processing and classification in a BCI system. A pilot study was conducted with 10 subjects to obtain preliminary results and evaluate the system's performance. This approach allowed for testing the viability of the proposed BCI paradigm and refining the data acquisition protocol. These initial findings guide adjustments to the experimental setup, ensuring the system is optimized before scaling up to involve a larger group of participants for extensive data collection. The results underscore the potential of integrating BCI technology with edge AI platforms for real-time processing. However, the findings also reveal significant challenges, particularly the highly unbalanced datasets in multi-class classification tasks. Moreover, the study highlights the need to shorten trial lengths to improve the detection of ErrPs. Addressing these challenges is essential for refining the experimental protocol and optimizing system design before scaling to larger datasets and broader subject participation. Overall, the study demonstrates the promise of edge-based BCI systems in enabling low-latency, energy-efficient, and scalable assistive devices. This approach has the potential to significantly enhance the autonomy and safety of individuals with severe motor disabilities, providing a practical and accessible solution for assistive robotic applications.