| dc.description.abstract | In some branches of work, such as operating rescue robots during a disaster, a badly adjusted workload can result in errors with devastating consequences. If we could estimate the workload the operator is experiencing, we could better assign the amount of work among the operators, and so optimize performance. Existing approaches compute a cognitive taskload (CTL), which maps the cognitive resources required for a set of tasks, to a numeric value. To improve this approach, we wish to extend it with a component modeling the emotional state (ES) in terms of arousal and valence, from the operator's physiological measurements.
One experiment was set up with young children in a relaxed setting, where we measures heart rate, analyzed video footage with facial activity detecting software and manually annotated smiling and frowning as a type of ideal sensor. Another experiment was set up with adults performing cognitively demanding tasks, where we measures galvanic skin response, heart rate and activity of the corrugator (frowning) and orbicularis oculi (smiling) muscles.
None of the collected measurements were sufficiently accurate to detect a difference between the most and least cognitively demanding or exciting aspects of the sessions. However, when using the manual annotations of smiling, the fuzzy logic model computed valence with an average accuracy of 95%, and a minimal accuracy of 80%. We also discovered a Pierson's correlation between frowning and cognitive taskload of 0.9.
Assuming that the future will bring an improvement in accuracy of physiological measuring devices, fuzzy logic offers a simple, transparent, fast way of modeling an emotional state. | |
