EEG-Driven Intention Decoding: Offline Deep Learning Benchmarking on a Robotic Rover

Ghadah Alosaimi; Maha Alsayyari; Yixin Sun; Stamos Katsigiannis; Amir Atapour-Abarghouei; Toby P. Breckon

cs.ROFeb 23, 2026

EEG-Driven Intention Decoding: Offline Deep Learning Benchmarking on a Robotic Rover

Authors: Ghadah Alosaimi, Maha Alsayyari, Yixin Sun, Stamos Katsigiannis, Amir Atapour-Abarghouei, Toby P. Breckon

Organizations: Department of Computer Science, Imam Mohammad Ibn Saud Islamic University, Saudi Arabia · Department of Computer Science, Durham University, UK · Department of Computer Science, King Saud University, Saudi Arabia · Department of Engineering, Durham University, UK

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 commands forward, reverse, left, right, and stop. Electroencephalogram (EEG) signals were recorded with a 16-channel OpenBCI cap and aligned with motor actions at Delta = 0 ms and future prediction horizons (Delta > 0 ms). After preprocessing, several deep learning models were benchmarked, including convolutional neural networks, recurrent neural networks, and Transformer architectures. ShallowConvNet achieved the highest performance for both action prediction and intent prediction. 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 deep learning-based BCI systems.

Figures & tables

Fig. 1 : Experimental setup for EEG-based driver intention prediction. EEG signals were acquired from the subject and synchronised with a first-person view of the route. Commands were sent via a controller and recorded for data labelling. The rover was equipped with a ZED stereo camera, GNSS antenna, Lux, and IMU sensors to capture multimodal information during the experimental route. The recorded data were processed through preparation and labelling pipelines before being used for driver intention decoding.

Family	Model	Purpose	Focus	Strengths
CNN	EEGNet [ 18 ]	Compact EEG decoding	Spatial-temporal features	Robust to noise; lightweight
	DeepConvNet [ 33 ]	Deep feature extraction	Spatial-temporal features	Strong representation learning
	ShallowConvNet [ 33 ]	Short-window EEG decoding	Temporal patterns	Interpretable filters; effective for MI
	STNet [ 41 ]	Multi-channel EEG modelling	Spatial-temporal dependencies	Captures electrode interactions
	TSCeption [ 8 ]	Frequency-aware EEG decoding	Temporal-spectral features	Multi-scale kernels
	CCNN [ 37 ]	Real-time EEG classification	Low-latency processing	Lightweight; portable BCI suitability

TABLE I : Deep learning models evaluated for EEG classification, grouped by architecture family.

Panel of Figure 3 — Fig. 2 : Aggregated confusion matrices for (a) ShallowConvNet, (b) GRU, and (c) EEGConformer, at $\Delta=0$ ms.

Panel of Figure 4 — Fig. 3 : Aggregated confusion matrices for (a) ShallowConvNet, (b) GRU, and (c) EEGConformer, at $\Delta=300$ ms.

		Action ( $\mathbf{\Delta=0}$ ms)				Intention ( $\mathbf{\Delta=300}$ ms )
Family	Model	Accuracy	Precision	Recall	F1	Accuracy	Precision	Recall	F1
CNN	EEGNet	0.78 $\pm$ 0.13	0.63 $\pm$ 0.11	0.70 $\pm$ 0.13	0.64 $\pm$ 0.12	0.77 $\pm$ 0.13	0.62 $\pm$ 0.10	0.68 $\pm$ 0.13	0.62 $\pm$ 0.12
	DeepConvNet	0.33 $\pm$ 0.17	0.35 $\pm$ 0.14	0.35 $\pm$ 0.10	0.24 $\pm$ 0.12	0.32 $\pm$ 0.16	0.35 $\pm$ 0.14	0.35 $\pm$ 0.10	0.24 $\pm$ 0.12
	ShallowConvNet	0.83 $\pm$ 0.11	0.67 $\pm$ 0.10	0.72 $\pm$ 0.12	0.67 $\pm$ 0.11	0.82 $\pm$ 0.11	0.65 $\pm$ 0.09	0.71 $\pm$ 0.12	0.66 $\pm$ 0.11
	STNet	0.61 $\pm$ 0.15	0.46 $\pm$ 0.13	0.47 $\pm$ 0.14	0.46 $\pm$ 0.13	0.61 $\pm$ 0.15	0.46 $\pm$ 0.13	0.47 $\pm$ 0.14	0.45 $\pm$ 0.13
	TSCeption	0.81 $\pm$ 0.11	0.62 $\pm$ 0.11	0.63 $\pm$ 0.12	0.62 $\pm$ 0.11	0.81 $\pm$ 0.10	0.63 $\pm$ 0.11	0.62 $\pm$ 0.11	0.62 $\pm$ 0.11

TABLE II : Model performance comparison at

\Delta=0

ms (action) and

\Delta=300

ms (intention).

Panel of Figure 6 — Fig. 4 : F1-scores across all labelling horizons ( $\Delta\in[0,1000]$ ms) for all examined models.

Explore similar work

Mind2Drive: Predicting Driver Intentions from EEG in Real-world On-Road Driving
Apr 21, 2026Ghadah Alosaimi, Hanadi Alhamdan, Wenke E +3Electroencephalography EEG Decoding
Department of Computer Science, Imam Mohammad Ibn Saud Islamic University, Saudi Arabia · Department of {2Computer Science, 3Engineering}, Durham University, UK · Department of Computer Science, Princess Nourah bint Abdulrahman University, Saudi Arabia
Robotic Grasping and Placement Controlled by EEG-Based Hybrid Visual and Motor Imagery
Mar 3, 2026Yichang Liu, Tianyu Wang, Ziyi Ye +4Human-Robot Interaction Robotic Grasping
Institute of Science and Technology for Brain-inspired Intelligence (ISTBI), Fudan University · School of Data Science, Fudan University · Shanghai Innovation Institute +3
Copilot-Assisted Second-Thought Framework for Brain-to-Robot Hand Motion Decoding
Mar 29, 2026Yizhe Li, Shixiao Wang, Jian K. LiuEEG Decoding Brain-Computer Interfaces
University of Birmingham, Birmingham, United Kingdom · University of Birmingham

Apr 21, 2026J/K move · Enter open · S save

Mind2Drive: Predicting Driver Intentions from EEG in Real-world On-Road Driving

Ghadah Alosaimi, Hanadi Alhamdan, Wenke E, Stamos Katsigiannis+2

Department of Computer Science, Imam Mohammad Ibn Saud Islamic University, Saudi Arabia · Department of {2Computer Science, 3Engineering}, Durham University, UK · Department of Computer Science, Princess Nourah bint Abdulrahman University, Saudi Arabia

Electroencephalography EEG Decoding Autonomous Driving

Open paper AskPDF arXiv

Predicting driver intention from neurophysiological signals offers a promising pathway for enhancing proactive safety in advanced driver assistance systems, yet remains challenging in real-world driving due to EEG signal non-stationarity and the complexity of cognitive-motor preparation. This study proposes and evaluates an EEG-based driver intention prediction framework using a synchronised multi-sensor platform integrated into a real electric vehicle. A real-world on-road dataset was collected across 32 driving sessions, and 12 deep learning architectures were evaluated under consistent experimental conditions. Among the evaluated architectures, TSCeption achieved the highest average accuracy (0.907) and Macro-F1 score (0.901). The proposed framework demonstrates strong temporal stability, maintaining robust decoding performance up to 1000 ms before manoeuvre execution with minimal degradation. Furthermore, additional analyses reveal that minimal EEG preprocessing outperforms artefact-handling pipelines, and prediction performance peaks within a 400-600 ms interval, corresponding to a critical neural preparatory phase preceding driving manoeuvres. Overall, these findings support the feasibility of early and stable EEG-based driver intention decoding under real-world on-road conditions. Code: https://github.com/galosaimi/Mind2Drive.

Abstract

Explore similar work

Mind2Drive: Predicting Driver Intentions from EEG in Real-world On-Road Driving

Robotic Grasping and Placement Controlled by EEG-Based Hybrid Visual and Motor Imagery

Copilot-Assisted Second-Thought Framework for Brain-to-Robot Hand Motion Decoding