Time estimation and error feedback in predictive eye movement timing

This dissertation investigates predictive motor timing in normal human subjects by monitoring rapid eye movements (saccades) to alternating targets. The research goal is to understand further and model how error feedback is used to pre-program motor responses. In general we find that targets alternating at low pacing frequencies promote reactive tracking (the saccade response lags behind the target jump) and high pacing frequencies promote predictive tracking (the saccade response anticipates the target jump). In the following chapters we demonstrate that these tracking behaviors exhibit different structures of time correlation: predictive tracking is influenced by a feedback process that takes into account the latency of previous trials whereas reactive tracking generates what appears to be a sequence of statistically independent values. In addition, we examine the effect of transient perturbations to the pacing stimulus during steady-state reactive and predictive tracking. Although reactive tracking was greatly affected by these stimulus perturbations, once predictive tracking was established subjects continued to time their saccade responses at the pre-existing rate despite the perturbations. This result suggests that predictive tracking is the result of an internal estimate of the stimulus timing. We conclude with a closed-loop timing model of predictive saccade tracking based on these behavioral results. In this model, the timing between movements is based on an internal estimation of stimulus timing which is represented by a linear rising signal to threshold, with added noise. The time estimation (the signal threshold of the integrate-to-fire mechanism) is adjusted by the feedback of the timing (inter-movement interval) and error (latency) of previous movements made within a time window of approximately 2 seconds. We apply the model to several experimental paradigms and the simulated results agree with the observed behavioral results. Using two related estimates of the window length, which were derived independently based on different sets of data, the model reconstructs the statistical structure of the predictive latency time series and reasonably reproduces the time window. The simulation confirms that predictive saccade tracking is based on an internal estimate of the stimulus timing modified by feedback of movements occurring within a temporal window of previous performance.