Figure 3.

Two models of feedforward motor planning. Arm position estimates X̂_INV are computed by combining visual and proprioceptive signals. A movement vector describing the desired direction of the initial velocity (Ẋ^*) or acceleration (ẍ^*) of the hand is computed by subtracting X̂_MV from the target location x^*. An inverse model transforms this desired extrinsic vector into an intrinsic motor command specifying joint angle velocities ( ) or torques (τ), depending on the model being implemented. This transformation makes use of a second position estimate, X̂ and X̂_INV. Finally, the motor command is executed, determining the initial hand trajectory. Note that the loop through the “Vision” and “Proprioception” boxes does not imply feedback control; in these models, the position estimates are only used to plan the plan the initial, feedforward component of a movement.