Real-time feasibility of online EEG-based intention decoding in mixed reality

Determine whether the computational requirements and end-to-end latency of online, real-time classification of user intention (Select versus Observe) from anticipatory EEG activity using deep learning models are compatible with mixed reality interaction, including the streaming preprocessing and inference necessary for immediate applicability in live systems.

Background

The paper demonstrates offline decoding of user intention (Select vs. Observe) from anticipatory EEG activity using several deep learning architectures, achieving high person-dependent accuracies and reporting low model inference times. However, these results were obtained offline and do not include end-to-end testing in an online, real-time mixed reality environment.

As acknowledged by the authors, evaluating online pipelines that include continuous EEG acquisition, streaming preprocessing, and real-time inference is necessary to establish practical viability for adaptive mixed reality interfaces. Specifically, the computational and latency characteristics of such online systems remain unverified, leaving open whether they meet real-time interaction requirements.