We used the hypothesis that rather than specifying torques, EMG signals or joint angles, control variables predetermine, in particular, the relationship (called the invariant characteristic or IC) between torque and position but not their individual values (Feldman, 1966; Levin, Feldman, Milner, & Lamarre, 1992). Figure 1 illustrates the present experiments in terms of ICs. The initial equilibrium point (EP) is shown by i and the initial joint IC by the dotted line. The three types of load characteristics are shown respectively by line L+ for those opposing wrist flexion (line rising to the left), L-for those assisting wrist flexion (line descending to the left) and line Lo for zero load (coinciding with abscissa). In Experiment 1 subjects trained to reach the target with no load by moving from EP i to a in the target zone (shaded area). This movement is believed to be produced by a shift in the IC of the joint to the left (line b-a-c). If the subject maintains the same IC by the instruction not to correct, the movement opposed by L+ should terminate at EP b undershooting the target. Thus, predicted point b which lies on the joint IC would be considered an error if the subject was instructed to correct the movement as soon as possible. Indeed, after the same initial training, the movement assisted by L-should terminate at EP c, overshooting the target. Similarly, in Experiment 2 when the subject trains to reach the target zone with an opposing load (EP e), he would shift the joint IC further to the
left (line e-d) and overshoot positional errors would be made to EP d in those trials that are randomly not loaded. On the other hand, if the subject trains to reach the target with the assisting load (EP g) thus arresting the shift in the joint Ie earlier Gine f-g), predicted movement errors would be undershoots to EP f in the random trials, in which the wrist was not loaded.