Alternative Indices of Performance: An Exploration of Eye Gaze Metrics in a Visual Puzzle Task

摘要

Of interest to the U.S. Air Force is the ability to develop and characterize the level of workload that operators are under at any given point. When an operator\u27s cognitive resources exceed demands, a \u27red line\u27 of performance may be crossed after which performance breaks down. What is needed is an estimate of operator state; a \u27dipstick\u27 for the operator in order to assess the level of \u27resources\u27 available, in order to avoid performance problems. Traditional approaches use secondary tasks (e.g., mental arithmetic) or secondary physiological measures (e.g., heart rate variability) for state assessment. However, the current work was motivated by dynamic systems theory which indicates that there are meaningful patterns of variability in \u27primary\u27 behaviors (e.g., required activities) which might provide a measure of operator state. The present work uses eye gaze as a primary measure in a visual puzzle task. The link between eye gaze and attention is generally accepted as is the link between attention and performance outcomes. The goal of Experiment 1 was to determine if performance changes in a visual puzzle task were reflected in eye gaze, as measured in multiple ways: conventional (e.g., average fixation length) and dynamic (e.g., Beta values, measures derived from a recurrence matrix). These relationships were explored in relation to task difficulty, practice, as well as spare capacity. The results of Experiment 1 suggest that there are impacts of task demands on gaze patterns, for both conventional and dynamic gaze metrics. There were also significant of practice on eye gaze patterns in Experiment 1 that could be interpreted as learning or strategy shifts. The impact of learning on eye gaze was explored in a follow up experiment. The results of Experiment 2 show a significant improvement in performance in the task accompanied by change in gaze patterns when repeating the same puzzle; and that the dynamic measure of diagonal recurrence was systematically related to this performance change. This suggests that non-conventional measures of dynamic structure provide additional, complimentary information about operator state.