Fig. 2

Real-time Control Flowchart. EMG data (top left) from the 8 implanted electrodes (with upper arm percutaneous lead exit sites shown behind) is collected by the Ripple Grapevine Neural Interface Processor (shown top center behind the signal processing text), which applies a Butterworth (BW) bandpass filter and digitally samples the signal (ADC: Analog-to-Digital Converter). Two features (Mean Absolute Value and Waveform Length) are then calculated from a 200 ms window of this EMG data every 50 ms and subsequently fed to a decoder. This regressive K Nearest Neighbor (KNN) controller computes a velocity vector for a virtual reality (VR) hand posture matching program (shown in the bottom right drawing). The gain (relationship between predicted and outputted virtual hand motion) and velocity threshold (below which all velocities are considered erroneous and set to zero) are then adjusted according to participant preferences. The resulting VR 3-DOF hand position is shown in the bottom left. The purple rectangles mark the target hand posture ranges for each DOF, with the black lines indicating the VR hand’s relative positions in those DOFs. The longer the purple rectangle, the more time the participant took to reach and keep that target hand posture for the required dwell time of 1 s