Extended Data Fig. 6: Demonstration of TFRFP-implanted FEHs for intelligent prostheses application.
a, Representative action-track vectors of the foot joints, and the corresponding EMG signals and spectrograms recorded via four-channel FEHs. b, Representative action signals recorded by a single-channel FEH. Eight typical action samples are presented, each comprising a foot joint action trajectory vector alongside corresponding EMG signals and spectrograms. c,d, Unsupervised learning results of the CNN PCA cascade machine learning model under the multi- (c) and single-channel mode (d). e,f, Representative results demonstrating the application of single- (e) and four-channel FEHs (f) coupled with the AI model in action track prediction and robotic arm control. Several representative testing results show the output-predicted action trajectory compared with the actual action trajectory. g,h, Ankle endpoint (g) and action vector angle deviation statistics (h) between AI-predicted results and the actual action track (n = 6 testing samples for four-channel FEHs and n = 10 testing samples for single-channel FEHs).
