[Paper Review] Towards Brain-Computer Interfaces for Drone Swarm Control

Ji-Hoon Jeong, Dae-Hyeok Lee+2 more|arXiv (Cornell University)|Feb 3, 2020

EEG and Brain-Computer Interfaces24 references4 citations

TL;DR

This study presents a noninvasive EEG-based brain-computer interface (BCI) for controlling drone swarms using visual imagery paradigms. It demonstrates feasible classification of four high-level swarm commands—'Hovering', 'Splitting', 'Dispersing', and 'Aggregating'—achieving a grand average accuracy of 36.7% across seven subjects, exceeding the 25% chance level, thus proving the viability of EEG-based drone swarm control with basic machine learning.

ABSTRACT

Noninvasive brain-computer interface (BCI) decodes brain signals to understand user intention. Recent advances have been developed for the BCI-based drone control system as the demand for drone control increases. Especially, drone swarm control based on brain signals could provide various industries such as military service or industry disaster. This paper presents a prototype of a brain swarm interface system for a variety of scenarios using a visual imagery paradigm. We designed the experimental environment that could acquire brain signals under a drone swarm control simulator environment. Through the system, we collected the electroencephalogram (EEG) signals with respect to four different scenarios. Seven subjects participated in our experiment and evaluated classification performances using the basic machine learning algorithm. The grand average classification accuracy is higher than the chance level accuracy. Hence, we could confirm the feasibility of the drone swarm control system based on EEG signals for performing high-level tasks.

Motivation & Objective

• To develop a brain-swarm interface system for noninvasive control of drone swarms using EEG signals.
• To investigate whether high-level swarm commands can be decoded from EEG during visual imagery tasks.
• To evaluate classification performance of four distinct drone swarm behaviors using basic machine learning.
• To establish a controlled experimental environment for EEG acquisition during simulated drone swarm operations.
• To assess the feasibility of EEG-based BCI for complex, multi-command drone swarm control in realistic scenarios.

Proposed method

• Employed a visual imagery paradigm where subjects imagined four drone swarm behaviors: Hovering, Splitting, Dispersing, and Aggregating.
• Used a 4-phase experimental protocol: rest, visual cue/preparation, fixation, and imagination, with a 90 cm monitor at eye level.
• Acquired EEG signals using a 64-channel EEG system at 1,000 Hz sampling rate with a 60 Hz notch filter and impedance <10 kΩ.
• Preprocessed data with a zero-phase 2nd-order Butterworth band-pass filter (8–30 Hz) and segmented into 4-second epochs.
• Applied Common Spatial Pattern (CSP) to extract spatial features, using logarithmic variances of the first and last three CSP components.
• Classified four classes using Linear Discriminant Analysis (LDA) with a one-versus-rest strategy and 5-fold cross-validation.

Experimental results

Research questions

• RQ1 Can EEG signals during visual imagery of drone swarm behaviors be reliably classified into four distinct command classes?
• RQ2 Does the classification accuracy of high-level drone swarm commands exceed the chance level (25%) using noninvasive EEG and basic machine learning?
• RQ3 How do individual differences in EEG signal quality and task performance affect classification accuracy in a BCI-based drone swarm control system?
• RQ4 To what extent does a simulated, realistic environment enhance the feasibility of EEG-based control for complex swarm tasks?
• RQ5 Can a brain-swarm interface be developed using only noninvasive EEG for multi-command, high-level control of drone swarms?

Key findings

• The grand average classification accuracy across all seven subjects was 36.7% (±4.6%), significantly exceeding the 25% chance level for four-class classification.
• Subject 5 achieved the highest accuracy at 41.3%, while Subject 3 recorded the lowest at 28.4%, indicating variability in individual EEG signal responsiveness.
• The EEG data quality was sufficient for reliable classification under controlled conditions, as evidenced by performance above chance level.
• Subject 3’s lower performance was linked to difficulty in performing the visual imagery task, suggesting cognitive load or task comprehension issues.
• The results confirm the feasibility of using EEG-based BCI for high-level drone swarm control, particularly for fundamental swarm behaviors.
• The study lays a foundation for future integration of deep learning to improve classification robustness in real-world environments.