200 likes | 318 Views
Sponsored by US Army Research Office SBIR Phase II: “Wearable Physiological Sensor Suite For Early Detection Of Cognitive Overload” US Army Aberdeen Test Center: “Remote Neurological Monitoring Program”. Stress, Overload, and Performance. Mental State Estimation. Statisitcal Relationships.
E N D
Sponsored by • US Army Research Office SBIR Phase II: “Wearable Physiological Sensor Suite For Early Detection Of Cognitive Overload” • US Army Aberdeen Test Center: “Remote Neurological Monitoring Program”
Mental State Estimation Statisitcal Relationships
Definitions • Engagement: selection of a task as the focus of attention and effort • Workload: significant commitment of processing resources to an engaged task • Visual, Auditory, Haptic • Psychomotor • Cognitive (memory, executive) • Overload: task demands outstrip available processing resources • Mental Fatigue: desire to withdraw attention and effort from an engaged task associated with extended performance (~45 min)
Fatigue-Related EEG Sources Black = Alert Red = Mentally Fatigued Parietal Fz Pz Alpha Frontal Theta
Experimental Controls • Task learning • Time of day and time on task • Test day • Food consumption • Neurotoxic effects • Test environment • Inadequate measurement of physiological variance • Inadequate definition of ground truth workload levels: • Expert analysis and scoring of replayed videos • Logging all user inputs • Measuring reaction times to probes
Validation of Workload Manipulation NASA - TLX questionnaires P300
Summary • Biosignals exhibit high sensitivity to mental states, such as engagement, workload, and fatigue • Accurate biosignal-based models or “gauges” can be developed under controlled conditions and extended to new conditions • However, cognitive gauges are not very stable over time, due to behavioral, strategic, and physiological variability • Multimodal models capture a wide range of behavioral and physiological variability, improving robustness of gauges over time and conditions • Signal processing and computational methods help, but are not enough to yield stable models • Some recalibration or model adaptation is currently required • We seek ways to stabilize models with a minimum of recalibration