Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{199965,
        author = {Saptarshi Majumder and Mohendra Dey and Shibangi Khan and Soujanya Sarkar and Diptargha Pattanayak and Anik Sarkar and Amit Mondal},
        title = {HYBRID DEEP LEARNING MODELS FOR REAL TIME EEG SIGNAL CLASSIFICATION IN NON-INVASIVE BCI CONTROLLED ASSISTIVE ROBOTICS},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {12},
        pages = {1308-1316},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=199965},
        abstract = {Non-invasive brain-computer interfaces (BCIs) relying on electroencephalography (EEG) still offers the most accessible pathways for translating neural intent into robotic action. A significant challenge persists: muscle artefacts, ocular movements, and environmental noise hinder progress. Most pipelines simplify the challenge, reducing it to straightforward "thought-to-action" mapping, ignoring the genuinely non-stationary and messy nature of raw EEG. Many of the systems ignore the rapid change and messy raw EEG signals. To overcome the limitation, we tried to build a hybrid deep learning system that can easily and rapidly interpret motor imagery movements to control assistive robotics. This model uses neural network layers to capture spatial and temporal patterns where instead of sending the raw data, it is being cleaned using techniques like wavelet filtering, subject specific artefact removal which are much more important than expected. The whole system gives improved data quality for the user being present in any situation, it can continuously monitor user intentions without frequent recalibration. When tested on normal datasets and our own recordings, the system achieved 92% accuracy with response time under 150 ms. So as a result, it is to be mentioned that the robotic assistance has become more reliable and smooth. Concluding that this work will help in bringing the non invasive BCI technology closer to the real time use prioritizing the better handling of noisy, raw brain signals.},
        keywords = {Non-invasive brain-computer interface, EEG motor imagery, Hybrid deep learning, Depth-wise separable convolution, Bidirectional LSTM, Artifact subspace reconstruction, Adaptive wavelet thresholding, Real-time assistive robotics.},
        month = {May},
        }