How Can the Trust-Change Direction be Measured and Identified During Takeover Transitions in Conditionally Automated...

Area of Science:

• Human-Computer Interaction
• Automotive Engineering
• Cognitive Psychology

Background:

• Automated driving systems (ADS) present challenges in maintaining driver trust during takeover transitions (TTs).
• Inappropriate reliance on ADS, including misuse and disuse, can arise from fluctuating driver trust.
• Understanding trust dynamics is crucial for safe human-AV interaction.

Purpose of the Study:

• To propose an objective method for measuring and identifying trust-change directions during TTs in conditionally automated vehicles (AVs).
• To explore the relationships between physiological responses, takeover factors, and trust changes.
• To develop machine learning models for recognizing trust-change directions.

Main Methods:

• Utilized a driving simulator with 34 participants experiencing takeover events.
• Combined unsupervised learning and statistical analyses to examine physiological data (skin conductance, heart rate) and takeover factors.
• Applied machine learning techniques, including random forest (RF), to build trust-change recognition models.

Main Results:

• Subjective trust ratings and monitoring behavior reliably indicated trust-change directions.
• Physiological parameters showed a negative correlation with trust-change directions.
• Factors such as longer takeover request (TOR) lead time, higher takeover frequency, and stationary vehicle scenarios were associated with increased driver trust.
• The RF model achieved an F1-score of approximately 77.3% in identifying trust-change directions.

Conclusions:

• The investigated features and the developed RF model accurately identify trust-change directions during TTs.
• These findings support the development of trust monitoring systems for AVs.
• Mitigating driver overtrust and undertrust is essential for safe operation of conditionally automated vehicles.