Cognitive Workload
Cognitive workload (CWL) refers to the mental demands imposed by a task relative to an operator's available cognitive capacity. It is distinct from physical workload and from arousal: it specifically captures the load placed on limited cognitive resources including working memory, attention, and executive function.
Why Workload Matters
Many high-stakes professions operate under conditions of sustained and highly variable cognitive demand. Operators may simultaneously track multiple dynamic entities, communicate across channels, coordinate with team members, anticipate future developments, and issue timely actions to resolve emerging problems.
Cognitive overload - when demand exceeds capacity - increases error probability dramatically. Underload (too little activity) also impairs performance through complacency and reduced vigilance. Optimal performance occurs within a moderate workload band.
NASA Task Load Index (NASA-TLX)
The NASA-TLX is the most widely used and validated subjective workload assessment instrument. It measures workload across six independent dimensions:
| Dimension | Abbreviation | What It Captures |
|---|---|---|
| Mental Demand | mental |
Amount of mental and perceptual activity (thinking, deciding, calculating, remembering, searching) |
| Physical Demand | physical |
Amount of physical activity (controlling, activating, pushing, pulling) |
| Temporal Demand | temporal |
Time pressure; pace of work; feeling rushed |
| Effort | effort |
How hard (mentally and physically) you had to work to achieve your level of performance |
| Performance | performance |
Perceived success in accomplishing the task goal (inverted: higher is lower workload) |
| Frustration | frustration |
Feelings of irritation, stress, annoyance, or insecurity |
Each dimension is rated on a 1–21 scale. The composite NASA-TLX score is the weighted or unweighted mean of the six dimensions.
NASA-TLX labels in this project
Most datasets in this project (UNIVERSE, WAUC, MOCAS) collect NASA-TLX scores after each experimental section. These scores serve as the primary regression or classification labels for workload decoder fine-tuning.
Physiological Correlates
EEG
| EEG Feature | Direction with High Workload | Neural Basis |
|---|---|---|
| Frontal theta power (F3, F4, Fz) | Increases | Prefrontal working memory load |
| Occipital/parietal alpha power | Decreases (alpha suppression) | Increased visual/attentional processing |
| Frontal midline theta (Fz) | Increases | Executive control and task engagement |
| P300 amplitude | Decreases | Reduced cognitive resources for oddball processing |
| N2 amplitude | Increases | Error monitoring under load |
Alpha suppression is the most consistently replicated EEG workload indicator: as task demand increases, occipital alpha power (8–13 Hz) decreases, reflecting increased visual cortical processing.
Frontal theta is the second most reliable marker: prefrontal theta (4–8 Hz at F3/F4/Fz) increases during working memory-intensive tasks, reflecting sustained prefrontal engagement.
ECG / PPG
- Heart rate generally increases with workload.
- High-frequency HRV (0.15–0.4 Hz) decreases with increasing mental load, reflecting reduced parasympathetic tone.
- LF/HF ratio increases - sympathetic dominance.
Eye Gaze
- Fixation duration increases (more processing per fixation) under workload.
- Scan path entropy decreases (more focused attention, less exploration of peripheral information).
- Pupil dilation increases with mental effort.
Speech
- Fundamental frequency (F0) may increase under time pressure.
- Speech rate increases (rushing) or becomes irregular under very high workload.
- Jitter and shimmer increase under physiological stress.
Workload Operationalisation
For classification purposes, NASA-TLX scores are typically binned into three workload levels:
| Class | NASA-TLX Range | Description |
|---|---|---|
| Low | 1–35 | Comfortable; well within capacity |
| Medium | 35–60 | Moderately demanding; near optimal performance zone |
| High | 60–100 | Near or exceeding capacity; error-prone |
For regression, the raw composite NASA-TLX score (normalised to [0, 1]) is used as the continuous target.
Instantaneous Self Assessment (ISA)
ISA is a simpler real-time alternative to NASA-TLX: a 5-point scale (1 = under-utilised, 3 = comfortable, 5 = fully stretched) designed to be administered more frequently without disrupting task performance. It provides a coarser but more temporally dense workload label, used in MOCAS.
Applications
Live Cognitive Overload Monitoring
The primary operational application is a real-time traffic-light alert system:
- Green: workload within acceptable bounds.
- Amber: elevated workload; supervisor awareness triggered.
- Red: overload detected; immediate supervisor intervention prompted.
The alert is updated on a rolling basis from the sliding-window BFM predictions, smoothed with a temporal filter to avoid spurious alerts from brief spikes.
Task Distribution Optimisation
When one operator's workload is predicted to be high, the supervisor interface highlights the option to redistribute tasks to team members with lower predicted workload. The prediction leads the actual state by the BFM's temporal context window, providing the lead time needed for a proactive redistribution decision.
Separation Error Prediction
Critical errors are more frequent under high workload. The workload prediction feeds into a combined safety model alongside gaze data and task metrics for error probability estimation. See XAI: LASTS Framework for how these predictions are made explainable.