Figure 5.

Block diagram for the control model. The input to the system is the current best estimates of the state of the hand and the target position. The controller uses a minimum-jerk control law to convert these to a jerk signal (change in acceleration) to which motor noise is added before being applied to modify the motion of the hand. In the model presented here, the plant is trivial, because we assume a perfect kinematic controller. The plant simply integrates the jerk signal to change the acceleration of the hand. Delayed sensory estimates of the state of the hand are combined in an optimal way (through the Kalman gain matrix) with equivalently delayed internal estimates of the state of hand, and updated through the state update equation to generate a new estimate of the state of the hand at the next time step. The resulting sequence of state estimates are propagated forward through the state update equation, using the motor commands generated during the delay period, to generated ongoing estimates of the current state of the hand.