Fig. 3From: Stable, three degree-of-freedom myoelectric prosthetic control via chronic bipolar intramuscular electrodes: a case studyData Processing for K Nearest Neighbor Training. Training data (two EMG features from each of the eight muscles) was collected during the presentation of target VR hand postures. Note that, for clarity, only the normalized Mean Absolute Value (MAV) features are shown here as the thin blue lines, one for each muscle. The target VR hand posture was initially presented to the participant during the ‘Prepare’ phase, shown as yellow rectangles. At the start of the ‘Go’ phase, shown in green, the participant continuously attempted to move their phantom hand to the presented hand posture. After 2 s, a rest posture was presented and the participant relaxed during a ‘Rest’ phase- these are shown in sky blue. Two 3 DOF target hand postures are presented in this illustration, requiring (1) simultaneous supination, wrist extension, and hand closing (shown above the leftmost ‘Prepare’ and ‘Go’ phases) and (2) simultaneous pronation, wrist flexion, and hand opening (shown above the subsequent ‘Prepare’ and ‘Go’ phases). The ‘Assumed Intent’ (thick black line) was computed by normalizing the movement needed to reach the target by the average of the rectified EMG of the sections marked ‘Active Movement’ by the red line. The dotted purple line represents how the regressive KNN trained from all collected training data ultimately predicted the 3 velocity signals from the EMG featuresBack to article page