Three-Position Vehicle Control Based on Neural Interface Using Machine Learning

Fradkov, Alexander; Babich, Nickolay · 2025 · Crossref

DOI: 10.15622/ia.24.1.1

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of controlling a wheelchair using a non-invasive brain-computer interface (BCI) based on electroencephalogram (EEG) signals. The primary motivation is to assist individuals with motor impairments by enabling them to steer a vehicle through imagined movements. The authors focus on a three-position control scheme (left, stop, right) to simplify decision-making under uncertainty. A key contribution is the proposal of a new machine learning classification algorithm, the Fuzzy Almost Nearest Neighbors (FANN) method, designed to reduce the sensitivity of the classic k-Nearest Neighbors (kNN) algorithm to the choice of the parameter *k* and to lower computational resource requirements for low-power devices. The experimental setup involved five participants using a wireless 21-channel EEG headset (NeoRecCAP) to control a MET COMPACT 15 wheelchair via a laptop running Python-based software. The system processed EEG data in 2-second frames with a sampling interval of 4 ms. Signals were preprocessed using a Butterworth filter within the 8–30 Hz frequency band. Participants performed imagined hand movements corresponding to left or right turns, cued by auditory signals. The control logic utilized an error matrix to calculate the frequency of correctly recognized left and right intentions. If the difference between these frequencies exceeded specific thresholds (±0.2), a control signal was sent to the wheelchair; otherwise, the system remained in a "stop" state. The system incorporated online adaptation, where successful trials were used to retrain the classifier during the session. The study conducted over 15 trials across five subjects. Results demonstrated significant improvement in recognition accuracy through both system adaptation and user training. During the initial session, the recognition accuracy was approximately 40%. By the final control testing phase, accuracy increased to approximately 90%. The FANN algorithm was shown to maintain high classification accuracy while being less sensitive to the selection of *k* compared to standard kNN. The software implementation was optimized for resource-constrained environments, demonstrating feasibility for deployment on low-power microcomputers like the Orange Pi 3B. The significance of this work lies in the development of a robust, adaptive BCI control system that achieves high accuracy (~90%) for wheelchair steering without requiring specialized technical expertise for operation. The introduction of the FANN method offers a practical improvement over traditional kNN by simplifying parameter tuning and reducing computational load. This approach facilitates the practical application of BCIs in assistive technologies, allowing for real-time adaptation to individual users and enabling deployment on affordable, low-power hardware. Future work aims to expand control capabilities to include forward movement and braking, as well as further mathematical modeling of the adaptation processes.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-nvidia on 2026-08-10; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• physical impairment