Abstract
Correction on the basis of previous errors is paramount to sensorimotor learning. While corrections of spatial errors have been studied extensively, little is known about corrections of previous temporal errors. We tackled this problem in different conditions involving hand movements (HM), saccadic eye movements (SM) or button presses (BP). The task was to intercept a moving target (3 possible velocities) at a designated zone (i.e. no spatial error) either with the hand sliding a pen on a graphics tablet (HM), a saccade (SM) or a button press (BP) that released a cursor moving ballistically for a fixed time of 330 ms. The dependency of the final temporal error on action onset varied from "low" in HM (due to possible online corrections) to "very high" in the BP condition. We analyzed the lag-1 autocorrelation (acf(1)) of action onset and the dependency on previous errors to study how trial-by-trial corrections were made. In conditions SM and BP, acf(1) was not different from zero denoting an optimal correction, while subjects under-corrected (acf(1) >0) in the HM condition. Interestingly, in conditions SM and BP action onset did not depend on the previous temporal error, but it did in the HM condition. However, this dependency was clearly modulated by the duration of movement time as faster movements depended less on the previous actual temporal error. One explanation for how subjects corrected in SM, BP and HM involving fast movements would be that they used the predictive error (i.e. intended action onset minus actual action onset). We applied a Kalman filter to action onset and found, on average, larger [Rv1] Kalman gains in SM and BP conditions denoting larger changes of an internal model [Rv2] based on the predictive error. The type of temporal error that is used seems to depend on the prediction versus online control dimension.