Based on the linked source; automatically prepared and checked against the original report.
Approach
The proposed system uses electroencephalogram data and motor imagery involving right- and left-hand movements as control signals. The article describes simulated wheelchair navigation rather than reporting a deployed physical wheelchair. Its approach connects imagined movements with interface commands for brain-computer interface control.
Data preparation
Researchers used a pre-filtered dataset from an open-source EEG repository. They segmented recordings into 19 × 200 arrays to capture the onset of hand movements, and the data were acquired at a sampling frequency of 200 Hz. The supplied summary provides no additional participant or testing details.
Model and interface
The study proposes TFormerEEG, a Transformer-driven deep-learning architecture for classifying motor-imagery EEG. It also integrates a Tkinter-based interface for simulating wheelchair movements. The source describes this as functional and intuitive, but does not establish performance in a physical wheelchair or clinical environment.
Reported results
TFormerEEG achieved 93.04% test accuracy compared with listed machine-learning baselines, including XGBoost, EEGNet and EEG-Deformer. Its mean accuracy was 91.18% through stratified cross-validation. These are reported model results from the supplied research summary, not evidence of treatment effectiveness, real-world mobility performance or operational safety.
