Skip to main content
. 2018 Nov 5;12:751. doi: 10.3389/fnins.2018.00751

Figure 3.

Figure 3

Decoding methods. (A) The initial Kalman filter estimates fingertip kinematics based on summed threshold crossings from either the current 50 ms time bin or an optimally chosen time lag (dashed lines show 50 ms lag). (B) Finger motion was characterized in 1 DOF per finger group such that a position of 0 corresponds full extension and 100 corresponds full flexion. Intention estimation was applied to the online kinematics of the initial Kalman filter by taking incorrect decoder velocities (blue arrow) and flipping them (green arrow) to point toward the desired target (red circle), on-target velocities are set to zero. Intention estimation is only applied to estimate parameters for the ReFIT decoder, no knowledge of target locations is used during online control.