EEG-Driven Intention Decoding: Offline Deep Learning Benchmarking on a Robotic Rover
Ghadah Alosaimi, Maha Alsayyari, Yixin Sun, Stamos Katsigiannis, Amir Atapour-Abarghouei, Toby Breckon
AI summary
Problem
Prior BCI driving studies are limited to simulated environments, single-command decoding, or lack systematic deep learning benchmarking, leaving a gap in real-world, multi-command, predictive intention decoding.
Approach
The authors conducted a real-world experiment where 12 participants remotely controlled a robotic rover while EEG was recorded, then benchmarked 11 deep learning models across CNN, RNN, and Transformer families for offline classification of five driving commands across multiple future prediction horizons.
Key results
- Real-world multi-command EEG dataset from outdoor rover navigation
- Temporal-stratified evaluation pipeline to prevent data leakage
- ShallowConvNet achieved highest F1-scores (67% immediate, 66% at 300ms prediction)
- Compact CNNs consistently outperformed RNNs and Transformers
Why it matters
Provides a reproducible benchmark and practical design insights for developing predictive, deep learning-based brain-robot interfaces for real-world navigation and shared autonomy.
Abstract
Brain–computer interfaces (BCIs) provide a hands-free control modality for mobile robotics, yet decoding user intent during real-world navigation remains challenging. This work presents a brain–robot control framework for offline decoding of driving commands during robotic rover operation. A 4WD Rover Pro platform was remotely operated by 12 participants who navigated a predefined route using a joystick, executing the following commands: forward, reverse, left, right, and stop. Electroencephalogram (EEG) signals were recorded with a 16-channel OpenBCI cap and aligned with motor actions at ∆= 0 ms and eight future prediction horizons (∆> 0 ms). After data preprocessing, eleven deep learning (DL) models were benchmarked for the task of intent classification, across the Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and Transformer architectural families. Shal- lowConvNet achieved the highest performance for both action prediction (F1-score 67% at ∆= 0 ms) and intent prediction (F1-score 66% at ∆= 300 ms), maintaining robust perfor- mance at future horizons. By combining real-world robotic control with multi-horizon EEG intention decoding, this study introduces a reproducible benchmark and reveals key design insights for predictive, DL-based BCI systems.