Stability and variablity in a rhythmic task behavioral data and dynamic models /

by Wei, Kunlin.

Using the perceptual-motor skill of rhythmically bouncing a ball as an experimental vehicle the present dissertation examined questions about control strategies and their acquisition, adaptation and transfer. Previous studies had already documented that the actor is sensitive to the stability properties of the task dynamics and performs the rhythmic actions with a strategy where effects of perturbations converge back to steady state without requiring error-correcting racket movements. This behavior is consistent with the predictions from stability analyses of a dynamic model of the task. Experiment 1 continued to scrutinize this prediction by applying a range of perturbation magnitudes, designed to be within and outside of the model’s basin of attraction. Results showed that even small perturbations that were predicted to equilibrate passively were blended with active control flexibly responding to perceived errors. However, the time course of return to steady state performance was qualitatively consistent with predictions from passive stability. Experiment 2 investigated how the actor combined passive stability with active control when the dynamic stability of the task was manipulated by varying the coefficient of restitution at the racket-ball contacts. To quantify the degree of control the covariance structure of the state variables was compared with model predictions. These predictions were obtained from a model that was extended by stochastic components to yield predictions about the structure of fluctuations at steady state. Results revealed that, paradoxically, variability of performance decreased with decreasing stability, contrary to common expectations in motor control. This was explained by increasing compensatory variability in execution, a signature of control. Hence, actors rely on passive stability when the stability of the system is high and employ more active control when stability is reduced. Applying the same variability and stability analysis Experiments 3 and 4 revisited issues of acquisition, adaptation and transfer in the same skill. Both experiments clearly demonstrate that the performance improvement is correlated with increased sensitivity to passive stability. Variability was evaluated in a space spanned by execution and result variables by applying the TNC-decomposition of variability (Tolerance, Covariation, Noise). Results highlighted how learning and adaptation is a migration through the execution space combined with the fine-tuning of covariation between relevant variables and a reduction of noise. Sensitivity to passive stability forms some abstract knowledge that is easily transferable across effectors. Compared to the mere examination of outcome measures, the decomposition method provided finer-grained insights about learning, adaptation, and transfer. Taken together, the four studies extend our understanding about how coordinated performance is achieved by exploiting task stability and active control accompanied by changes in the structure of variability. iii