Abstract
Humans recalibrate the mapping between their visual and motor systems when they perceive systematic changes in the environment. Two main factors influence the rate of this recalibration, the extent to which the current mapping is reliable (mapping uncertainty) and the extent to which the visual feedback is reliable (feedback uncertainty). As an optimal adaptor, the Kalman filter takes these factors into account and hence, may be best suited to model such a system. This model makes different predictions depending on the nature of the feedback noise. For correlated noise (Random walk), the mapping randomly shifts with every trial, which should increase the mapping uncertainty and therefore increase the adaptation rate. On the other hand, uncorrelated noise in the feedback (Gaussian noise around a constant mapping) should increase feedback uncertainty and therefore decrease the adaptation rate. To test these predictions for the visuo-motor system, we used a rapid pointing task similar to the one used by Burge, Ernst & Banks (2008). We also systematically varied the trial times over which subjects could learn the statistics of the environment. We replicated the result for random walks and found that adaptation was indeed faster. The expected decrease in the adaptation rate for the uncorrelated Gaussian noise, however, could not be found. Surprisingly, we did not find any significant effect of prolonged learning time on the adaptation rate in either noise environment. Hence, it appears that the temporal window for the estimation of the statistics underlying the visuomotor mapping is relatively small. The results further indicate, that the assumption the Kalman filter makes about the stationary statistics of its measurement and system noise may not be accurate. Hence, in future work it might be useful to alter the model in order to account for dynamic statistics of the measurement and internal noises.