Improving Response Times in Non-Invasive Brain-Computer Interfaces (BCIs) for Spatial Computing Environments

Abstract

The continuous improvement and enhancement in spatial computing devices is unquestionable, but the present use of non-intrusive Brain-Computer Interfaces (BCIs), including electroencephalography (EEG), is still substantially hindered by the problem of high latency that leaves the brain-computer interaction lagging behind perception. Experiencing more than 250 milliseconds of latency, this condition results in an incongruity of senses that lowers the quality of the user experience. This article introduces a dual processing system that merges spatially efficient, low-powered Transformer models with the Edge Computing paradigm. With the help of 50 users in a mixed-reality setting, it was established through the experiment that the suggested system cut down the overall latency time to an average of 85 ms, and at the same time, it was possible to achieve a motor imagery classification accuracy of 92%. The results here represent a crucial step towards the smooth adoption and incorporation of real-time BCIs into spatial computing systems.