Stability and variability in performance and learning of a rhythmic task

Dagmar Sternad
Departments of Kinesiology and Integrative Biosciences, Pennsylvania State University

     Full text: Not available
     Last modified: April 28, 2007
     Presentation date: 08/11/2007 9:15 AM in MCC
     (View Schedule)

Abstract
As in walking and running the skill of rhythmically bouncing a ball in the air is profoundly governed by its dynamics, where intermittent collisions are intertwined with continuous smooth movements. In previous work a mathematical model of the task, the ball bouncing map, was developed that afforded stability analyses with predictions about performance consistent with dynamically stable behavior. A series of experiments showed that human subjects performed the task consistent with the model predictions and exploited dynamic stability, i.e., they used a strategy where small noise or perturbations are compensated for without explicitly correcting for the error. For our recent experiments we created a virtual set-up with a haptic interface that afforded further tests of model predictions. The first experiment investigated the response of actors to perturbations, which was compared with simulation results of the model?s basin of attraction. Experimental data revealed a mixed control strategy showing that humans exploit dynamic stability but also actively adapt the racket movements to perceived perturbations. The second experiment investigated how the actor?s strategy changed when the contact dynamics was manipulated to produce different levels of stability. The deterministic ball bouncing model was extended with a stochastic component that yielded finer-grained predictions for steady state performance for different levels of stability. Results showed that with decreasing stability active control becomes more pronounced. The third and forth experiments investigated how novices learnt and adapted the skill. The analysis of dynamic stability was complemented by an analysis of variability which decomposed the dispersion in a set of data into three components: tolerance, covariation and noise (TCN). Variability analysis of the redundant task afforded quantification of the exploratory component next to compensatory strategy and reduction of stochastic noise during adaptation and learning. The three studies highlighted that humans flexibly intertwine different control strategies, mixing passive stability with active adjustments to changing task demands.