BCIwiki (bciwiki.com) — A 2-block lightweight architecture brought end-to-end latency in real-time EEG gait decoding down to 70.5 ms, according to a preprint posted to arXiv on August 3, 2026. The authors say EEG-based closed-loop control has been limited by motion artifacts and low signal-to-noise ratios. Their design splits the task: a front-end PolyTVL module compresses 32-channel EEG into a compact representation, and a back-end LSTM decoder classifies four gait phases (Stand, Initiate, Execute, Terminate).
In closed-loop deployment, the v01 configuration (PolyTVL+LSTM) reached a 55.3% gait initiation success rate with Rex assistance and 52.7% volitionally. The authors add that the manuscript was accepted for publication at the 2026 IEEE International Workshop on Machine Learning for Signal Processing (MLSP 2026).