Online decoding of rat self-paced locomotion speed from EEG using recurrent neural networks

Accurate and reliable neural decoding of locomotion holds promise for advancing clinical applications such as rehabilitation and prosthetic control, as well as for understanding neural correlates of action. Recent studies have demonstrated successful decoding of locomotion kinematics across species in motorized treadmill settings. But efforts to decode locomotion speed directly and continuously in more natural contexts—where pace is self-selected rather than externally imposed—are scarce, and those that exist generally achieve only modest accuracy and require intracranial implants. Here, we introduce an asynchronous brain–computer interface (BCI) that processes a stream of cortex-wide, 32-electrode skull-surface EEG recordings (0.01 - 45 Hz) to decode the instantaneous speed readouts of a non-motorized treadmill during self-paced locomotion in head-fixed rats. Using recurrent neural networks, our decoding methodology achieves 0.88 correlation (0.78 R²) for speed, primarily driven by visual-cortex electrodes and low frequency (< 8 Hz) oscillations. Moreover, pre-training on a single recording session permitted decoding on other sessions for only the same rat, suggesting the presence of uniform neural signatures of locomotion that generalize across sessions, but fail to transfer across animals. Finally, we found that cortical states not only carry precise information about the current speed, but also about future and past dynamics—up to 1000 ms. Our approach may therefore provide a useful framework for the development of high-performing, non-invasive BCI applications for locomotion.