Figure 6.

Examples of actual and predicted movements using Kalman filter decoding. A) Actual vs. offline-predicted wrist position values from one cross-validation testing set. Predictions were made using 32 spatially-filtered signals (CSP Cardinal 67% removed). B) Actual (black line) and offline-predicted wrist position as a function of time. Red line indicates predictions made with the same CSP filters as used in part A. Grey lines show offline predictions when no spatial filtering was used. C) 2D wrist trajectories when the animal’s actual hand position controlled the cursor position in real time. D) Predicted wrist positions using the neural data collected during the movements plotted in C. E) Cursor trajectories during closed-loop control using the same spatial filters and Kalman filter decoding function that was used offline in part D. Note that during closed-loop control, decoded wrist position was used to control cursor velocity in real time. Trajectories are color coded by intended target and black dots indicate when the cursor hit the target.