Multi-Class Error-Related Potentials for Correcting Robot Navigation Mistakes: A Human-in-the-Loop Approach Using EEG Brain-Robot-Interfacing

This study explores the integration of passive Brain-Computer Interfaces (BCIs) into robotic navigation to enhance obstacle detection utilizing human cognitive feedback. A multi-class classification approach was employed to identify four levels of navigation errors based on Error-related Potentials (ErrPs): no error, small, medium, and big errors. EEG signals were collected from 10 participants using a custom-designed interactive video game, and then the signals were classified using Support Vector Machines (SVM). The system achieved high classification accuracies across the error levels: 92.6% for no error, 92.8% for small error, 90.5% for medium error, and 89.6% for big error, with an average accuracy of 90.5% across all subjects. The results indicate that the proposed framework effectively captures graded cognitive responses to navigation errors, highlighting its potential for improving robot decision-making in complex or sensor-constrained environments.