Reliable, objective estimation of cognitive workload has potential applications in training (e.g., facilitating curriculum development), human performance assessment (e.g., treating workload itself as a performance metric), the design and development of human-automation teaming systems (e.g., evaluating the impact of design choices on operators’ cognitive workload), and adaptive automation (i.e., adapting automation behavior based on the cognitive workload of human operators). A wide variety of physiological indicators of cognitive workload have been investigated over the past five decades, including heart rate/variability, respiratory measures, pupillometrics, electrodermal activity (EDA), and indicators extracted from complex sources such as functional near-infrared spectroscopy (fNIRS) and electroencephalography (EEG). However, individual physiological indicators are non-specific to workload and must be combined with others in order to derive a useful estimate. The sensitivity and specificity of joint estimates depend on the sensitivities of the individual indicators to variations in cognitive workload and the unique information contributed by each.
This paper explores the utility of face and neck surface electromyography (fnsEMG)—non-invasive, skin surface measurement of the motor action potentials that drive muscle activity—as a new sensing modality for cognitive workload and its associated emotional responses. The sensitivity of fnsEMG to cognitive workload variations at nine face and neck sensing locations was evaluated in a Defense Advanced Research Projects Agency (DARPA) funded human study in which participants performed multiple concurrent cognitive tasks in a modified version of the National Aeronautics and Space Administration (NASA) Multi-Attribute Task Battery (MATB). Task performance and frequent self-reports of task difficulty were compared with multiple physiological signals, including fnsEMG, electrocardiography (ECG), EDA, respiration, eye gaze, and pupil size. A machine learning algorithm was trained to generate well-calibrated predictions of task errors and self-reported task difficulty based on these physiological signals, demonstrating their combined sensitivity to cognitive load and overload.
Keywords
ASSESSMENT, COGNITIVE, HUMAN PERFORMANCE, MACHINE LEARNING, PHYSIOLOGICAL
Additional Keywords