Abstract
When we repeatedly hit a moving target, we can make adjustments based on prior errors (e.g. start moving later if we were early previously). Different types of random noise (e.g. in the movement initiation and motor execution) contribute to this error. Therefore, the internal programming of next movement's adjustment should discount part of the error. To test this hypothesis subjects had to intercept a target at a designated interceptive area (no spatial error) and analyzed the dependence of the time of action initiation (TAI) on the previous TAI and temporal error. We addressed the problem of estimating the internal temporal adjustments by fitting a Kalman filter (KF) to the time series based on the TAI. The estimated transition of hidden states in the KF identifies the internal adjustments with a process noise (w) and the observed TAI is the measurement that includes the propagated motor noise. The previous temporal error served as control input whose coefficient (B) provides an estimate of the fraction of the observed error that is effectively used to determine the new hidden state. For each subject and session we estimated w and B as free parameters of the KF. On average, B was 0.15: the internal adjustments of the TAI accounted for the 15% of the total temporal error in the previous trial. The process noise (w) was between 6 and 10 ms and corresponded to 26% of the TAI variability. Finally we simulated the internal adjustments as corresponding variations in the initial state of a diffusion process and the arrival times as the TAI. Variations of 15% in the initial state led to adjustments in TAI that corresponded to full corrections of the temporal error as observed in the experiment. The internal timing of timed actions seems then tuned to propagated motor noise.
Meeting abstract presented at VSS 2016