Discovering optical control strategies: A data-mining approach

A major focus of research into visually guided action (VGA) is the identification of control strategies that map optical information to actions. The traditional approach to this research has been to test the behavioral predictions of a few hypothesized strategies against subject performance in environments in which various manipulations to available information are made. While important and compelling results have been achieved with this standard approach, they are potentially limited by small sets of hypotheses and the methods used to test them. In this dissertation, I introduce a novel application of data-mining and machine-learning techniques in a comparatively "assumption-lite" analysis of experimental data that is able to both describe and model human behavior. This method identifies the signature of optical control in the information-action patterns that are embedded in subject data. I also introduce a modeling philosophy that conceptualizes continuous control as an effort to realize preferred trajectories through an optical state space. In applying this philosophy, I show how simple models of the evolution of these trajectories can be mined from subject data. I demonstrate the effectiveness of this approach in the analysis and modeling of data from a collision-avoidance task and in the development of a new model of visually guided, speed-controlled steering.