WIRELESS GRAPHENE ECOG REVEALS DISTRIBUTED MOTOR-CORTEX MOTIFS FOR DECODING NATURALISTIC ACTIONS

Natural behaviour is a continuous stream of locomotor, postural and orofacial actions, yet most neural decoding frameworks remain trial-based and head-fixed. We asked whether chronic, fully wireless graphene electrocorticography (gECoG) can support stable, interpretable decoding of composite natural behaviours during unconstrained movement. Two rhesus macaques were implanted with flexible gECoG arrays over dorsal frontal motor cortex and recorded with an untethered wireless headstage during free movement. Video was annotated for walking, sitting, turning and chewing, and neural activity was aligned to behavioural transitions. All animal procedures were performed under approved institutional protocols. We mapped behaviour-locked band-power modulation and quantified transition-centred, frequency-resolved separability, then trained a hierarchical wavelet-based decoder with channel-by-band relevance weighting. Natural actions showed reproducible motifs: locomotion/turning were marked by widespread α/β suppression, sitting by relative α/β increases, and chewing by focal lateral γ enhancement. Discriminative information peaked around transitions, with β dominating sustained state contrasts and γ contributing transient, boundary-locked information. Restricting inputs to the most informative α/β/γ features matched an all-band baseline (validation accuracy about 85% vs about 86%) while reducing training time by about 37%. Online, the multithreaded pipeline achieved 25 to 33 ms processing and enabled brain-synchronised control of a quadrupedal robot that mirrored action sequences (frame-wise similarity 0.69; Levenshtein similarity 0.74). These results establish a transition-centred, physiology-grounded framework for power-efficient decoding of natural primate behaviour using untethered wireless gECoG.