Investigating Methods for Cognitive Workload Estimation for Assistive Robots

Area of Science:

• Human-Robot Interaction
• Cognitive Science
• Physiological Computing

Background:

• Assistive robots require sensitivity to human cognitive states to provide timely support.
• Accurate detection of human cognitive workload is crucial for effective human-robot interaction.
• Current sensing modalities for inferring workload are not fully understood.

Purpose of the Study:

• To identify the most effective physiological sensing modality for inferring human cognitive workload.
• To compare the performance of various machine learning models in workload classification.
• To inform the design of future human-robot interactive systems.

Main Methods:

• Analysis of physiological signals (eye gaze, electroencephalography, arterial blood pressure) from a simulated driving study.
• Application of machine learning models (k-nearest neighbor, naive Bayes, random forest, support-vector machines, neural networks) for workload inference.
• Statistical analysis to determine the significance of different modalities.

Main Results:

• Eye gaze signals demonstrated the highest accuracy (80.45%) in binary workload classification using support-vector machines.
• Combining eye gaze with electroencephalography yielded lower accuracy (77.08%) with neural networks.
• Eye gaze proved to be a superior indicator of cognitive workload, even when integrated with other signals.

Conclusions:

• Eye gaze is the most reliable physiological indicator for estimating human cognitive workload in human-robot interaction.
• The ease of collection and processing of eye gaze data makes it ideal for real-time applications.
• Findings support the development of more adaptive and responsive assistive robotic systems.