Continuous Cognitive Workload Assessment and Combined Metrics of Performance in the Multi-Attribute Task Battery
Presented at the 11th International Conference on Applied Human Factors and Ergonomics Conference (AHFE), San Diego, CA (July 2020)
Neurophysiological correlates of cognitive workload (e.g., changes in brain blood oxygenation) are detectable with minimally-invasive wearable sensors including functional near infrared spectroscopy (fNIRS). When cognitive workload increases, there is a corresponding increase in prefrontal blood flow that correlates with increased task engagement. Once the task becomes too difficult, there is a decrease in blood flow that correlates with disengagement from the task and decreased performance (Bunce, Izzetoglu, Ayaz et al., 2011; Ayaz, Cakir, Izzetoglu et al., 2012). Less clear, however, is the nature of these correlations between workload and neural activity during more complex, ecologically-valid tasks. While recently, some work has attempted to analyze increasingly realistic tasks such as driving in a car simulator (Unni et al., 2017), open questions remain.
Specifically, we hypothesized that on ecologically-valid tasks with multiple subtask components, a global combined metric of performance will better capture cognitive effort, and as such will yield stronger correlations with neurophysiology than will metrics capturing only single subtask components. We used a multi-component battery designed to be more similar to real world tasks, the multi-attribute task battery (MATB), first released in 1992 (Comstock Jr and Arnegard, 1992), and revised in 2011 (Santiago-Espada, Myer, Latorella et al., 2011). It includes a system monitoring task, a tracking task, a communications task, and a resource management task. During completion of three pre-validated levels of low, medium, and high difficulty (Nelson, 2016), we collected performance data along with fNIRS data. Subjects completed the three levels of the MATB before and after a boredom induction task for a total of six MATB blocks (two at each difficulty level). While simple proxies of workload such as MATB difficulty level and individual subtask performance levels showed nonsignificant or weak correlations with prefrontal neurophysiological activity, a global metric aggregating performance across subtasks in a balanced manner was strongly correlated with prefrontal neurophysiological activity.
We also examined the temporal properties of subject behavior as the task difficulty increased, finding subjects’ error rates did not consistency increase with increased difficulty, but instead that subjects increasingly ignored subtask cues. Additionally, subject did not neglect specific subtasks while maintaining focus on other subtasks for the entire duration of the blocks. Rather, subjects seemed to miss cues evenly across the subtasks, providing an explanation for why any single subtask performance metrics was not highly correlated with neurophysiological data. Correlations were reduced across all individual subtasks as there were times where subjects were likely not attending to that subtask.
These results inform future efforts to quantify the relationship between performance and neurophysiology within the context of the MATB and other similarly complex tasks, and in so doing assist the transition of human performance and human-machine interface research findings out of the laboratory and into actual use.
This work was funded under Government Contract No. NNX15CJ17P, SBIR Topic #H12-02. The views, opinions and/or findings contained in this report are those of the authors and should not be construed as an official NASA position, policy or decision unless so designated by other documentation.
1 Charles River Analytics
2 SA Technologies
For More InformationTo learn more or request a copy of a paper (if available), contact Bethany Bracken.
(Please include your name, address, organization, and the paper reference. Requests without this information will not be honored.)