Detecting unacceptable behavior of an autonomous vehicle using electroencephalography

Area of Science:

• Neuroscience
• Human-Computer Interaction
• Autonomous Systems

Background:

• Autonomous vehicles (AVs) require intuitive human interaction for safety.
• Left-turn maneuvers through oncoming traffic are complex scenarios demanding precise human-AV coordination.
• Understanding human cognitive responses to AV behavior is crucial for developing trust and safety.

Purpose of the Study:

• To investigate human cognitive processing during simulated autonomous vehicle left-turn maneuvers.
• To identify neural correlates, specifically event-related potentials (ERPs), associated with incongruent AV behavior.
• To explore the potential of ERPs for detecting critical situations in human-AV interaction.

Main Methods:

• Electroencephalography (EEG) was recorded from 33 participants observing a simulated AV.
• Participants viewed the AV executing a left-turn maneuver through oncoming traffic.
• ERPs, including N1, N2, P2, and P3 components, were analyzed based on AV behavior congruence.

Main Results:

• A significant increase in the N2 component (251-431 ms) was observed when the AV's behavior was incongruent with participants' assessment of the situation.
• No significant differences were found in N1, P2, or P3 components between congruent and incongruent conditions.
• These findings indicate the N2 component's sensitivity to discrepancies in human-AV interaction during critical maneuvers.

Conclusions:

• Event-related potentials, particularly the N2 component, show promise for real-time identification of critical situations in human-AV interaction.
• ERP-based monitoring could enhance the safety of autonomous driving systems by detecting human-AV behavioral mismatches.
• Further research is necessary to translate these fundamental findings into practical applications for autonomous driving safety.