Embedded AI‐Enabled Wearable Piezoelectric Sensing Cluster for Real‐Time Multijoint Motion Recognition

ABSTRACT Wearable motion recognition holds significant promise in rehabilitation medicine and human–machine interaction. However, they face challenges like signal susceptibility to interference and limited back‐end processing capabilities. This work reports a novel embedded artificial intelligence‐enabled sensor cluster featuring high assembly flexibility, good anti‐interference ability, and sensor nodes that autonomously process data, enabling real‐time recognition of multiple joint movements. Polyvinylidene fluoride‐based membranes for piezoelectric sensors exhibit high piezoelectric properties due to interface enhancement mechanisms resulting from hot‐pressing and rapid annealing. The well‐designed differential structure enhances the signal‐to‐noise ratio of the piezoelectric sensor to 72.5 dB, outperforming other reported polymer‐based flexible piezoelectric sensors (31 dB). The single‐joint recognition system used to build the cluster is equipped with a 12‐channel sensor array and a miniaturized signal‐conditioning circuit, which can real‐time recognize 20 different joint movements via a lightweight convolutional neural network model deployed on a microcontroller. Finally, the distributed multijoint motion recognition cluster adopted a one‐master‐multiple‐slaves architecture and multipoint wireless collaboration to synchronously recognize motions of the wrist, elbow, and shoulder. This work provides guidance for constructing motion recognition systems based on piezoresistive, piezoelectric, capacitive, and triboelectric principles.