Modeling the dynamics of visuo-motor adaptation behavior with a Kalman Filter

When the introduction of a prism makes visually guided reaching biased and inaccurate, adaptation occurs to restore accuracy. Bias in everyday life usually accumulates from a series of small changes, a process well simulated with a random walk. If bias were the only source of error, recalibration would be simple: correct the last error. But reaching behavior is also subject to random error. How does the visuo-motor system balance the need to filter random error with the need to adapt to time-varying bias? We investigated whether the Kalman filter, the optimal algorithm for this problem, models the dynamics of visuo-motor adaptation. The filter predicts that adaptation rate will be determined by the relative variances of current measurements and changing bias: rate should decrease with feedback variance and increase with variance in bias. Subjects pointed rapidly with an unseen hand to a brief visual target. Visual feedback indicated the endpoint of the motor movement. Feedback variance was increased with blur. The relationship between visual feedback location and the movement endpoint was altered with a random walk. Trial-by-trial pointing was measured. Subjects performed with a high level of efficiency and responded to changes in relative variances as predicted by the Kalman filter