This study presents an EEG-based Brain–Computer Interface for intuitive robotic navigation driven by ocular activity. A real multi-subject dataset collected from 15 participants was used to extract blink- and gaze related EEG features for five control commands. Eight supervised classifiers were evaluated under stratified 5-fold cross-validation: Random Forest, multilayer perceptron, support vector machine, k-nearest neighbors, RUSBoost, Naive Bayes, decision tree, and linear discriminant analysis. Among them, Random Forest achieved the best overall performance, reaching 98.74 ± 1.19% accuracy, 0.9874 Macro-F1, and 0.9993 macro-AUC, demonstrating excellent robustness and class separability across the five ocular classes. The decoded commands were transmitted wirelessly to an embedded Raspberry Pi platform, where they were converted into safe motor actions for wheelchair type robot navigation. Real world experiments in indoor environments confirmed stable motion control, reliable command execution, and successful obstacle avoidance without physical interaction from the user. These findings support the feasibility of a low-cost, portable, and non-invasive BCI solution based on realistic multi-subject EEG data for assistive mobility applications. Although the full control loop operated online, the end-to-end system latency was not quantitatively benchmarked in the present study and remains an important limitation to be addressed in future work. Future developments will focus on expanding the command set, improving robustness under more variable conditions, and validating the system in broader assistive mobility scenarios.
Nechchad et al. (Sat,) studied this question.