Silent failure detection in partial automation as a function of visual attentiveness

Area of Science:

• Human-Computer Interaction
• Automotive Engineering
• Cognitive Psychology

Background:

• Level 2 automation allows drivers to disengage from the dynamic driving task but requires continuous monitoring.
• Drivers often engage in non-driving related tasks, creating a trade-off between automation reliance and situational awareness.
• Gaze behavior management is critical for safe transitions during partial driving automation.

Purpose of the Study:

• To model driver gaze behavior during Level 2 partial driving automation.
• To investigate the impact of engaging in an email task on driver attention.
• To analyze driver response to a silent automation failure scenario.

Main Methods:

• Utilized sequence analysis methodology, including clustering and Hidden Markov Models (HMMs).
• Developed predictive models for four distinct driver gaze behaviors.
• Conducted a driving simulator study using the NADS-1 high-fidelity motion-based simulator.

Main Results:

• Predictive simulations accurately estimated driver gaze response times.
• Actual glance response times fell within predicted distributions.
• Three drivers with the longest response times failed to regain control before a simulated collision.

Conclusions:

• Simulated glance response time distributions offer valuable data for future driving performance analysis.
• Gaze response time probabilities can be conditioned on various factors like time on task and driver behavior.
• Sequence analysis is a flexible tool for future studies on driver gaze behavior and automation safety.