Development of a human performance modeling tool for user interface evaluation in next generation spacecraft: an empirical approach

Computational models of human performance hold considerable promise for streamlining design/test/redesign cycles for crew interfaces on next-generation space vehicles. To fulfill that promise, however, the models will have to be able to simulate and predict human performance in a highly complex operational environment. We describe an empirical approach to aid model development, which is based on analyses of the oculomotor behavior of human operators during a fault management task in a part-task shuttle cockpit simulator. For each operator, each of the eye fixations involved in the fault management task was classified as belonging to one of nine information acquisition categories (e.g. reading text, checking systems parameters, monitoring an action). For each category, we combined the fixation data across the operators and two levels of automation to produce a well populated distribution of fixation latencies, each of which was well fit by a gamma function. We then performed simulations in which, for each operator, the duration of each fixation involved in the fault management task was repeatedly stochastically sampled from the appropriate gamma distribution. The average of these samples was then substituted for the actual duration of each fixation, and the sum of the model-based fixation durations was then compared to the actual fault management completion time obtained from each observer. The model-derived task completion times were highly predictive of both the mean and variance in operators' actual finishing times. The next steps in generating model-based behavior are discussed.