A Minimal Brain-Computer Interface for Efficient Mission-Level Control in Industrial Human-Robot Interaction

As robots increasingly operate in industrial environments, ergonomic and hands-free human-robot interaction becomes essential. This paper proposes a brain-computer interface (BCI) for mission-level icon selection (e.g., scan, mill, pick, return), emphasizing operator comfort and low computational cost. Electroencephalography data are preprocessed by low-pass filtering, and wavelet coefficients are extracted as descriptive features. To reduce hardware complexity, two electrode selection methods are tested: (i) Forward Stepwise Feature Selection and (ii) Independent Component Analysis based ranking. For classification, a compact neural network with Fuzzy C-means clustering is employed. Using a public P300 speller dataset, mission icons are mapped to the stimulus matrix to emulate drone mission selection in warehouse scenarios. The reduced-channel approach achieves accuracy comparable to full-channel models while requiring only six electrodes, a training set of 360 samples, and fewer than 10-16 training epochs. The results show that compute-efficient BCIs with minimal electrodes can effectively support practical mission-level robot control in industrial settings.