Driver Take-Over Behaviour Study Based on Gaze Focalization and Vehicle Data in CARLA Simulator

Area of Science:

• Human-Computer Interaction
• Automotive Engineering
• Cognitive Psychology

Background:

• The automotive industry is rapidly advancing towards autonomous vehicles, necessitating a study of the human-machine interaction during the transition phases.
• Current limitations in autonomous technology and legal frameworks require drivers to retake control, making the handover process critical for safety.

Purpose of the Study:

• To investigate driver behavior during the transition between autonomous and manual driving modes.
• To analyze driver gaze focalization and reaction times in response to takeover requests within a simulated environment.

Main Methods:

• Utilized the CARLA simulator for a novel take-over study, tracking driver gaze using a camera-based, non-intrusive method (OpenFace 2.0 toolkit and NARMAX calibration).
• Fused gaze data with road semantic segmentation to pinpoint driver attention.
• Employed a dual-computer system with the Robot Operating System (ROS) framework for simulator portability and flexibility.

Main Results:

• Presented transition analysis results for 20 users across two scenarios using established metrics.
• Introduced a novel metric to assess driver situation awareness during mode transitions.
• Quantified driver attention and reaction times during critical takeover events.

Conclusions:

• The study provides valuable insights into driver behavior during autonomous to manual transitions, essential for designing safer autonomous systems.
• The developed gaze tracking method offers a cost-effective and non-intrusive approach for driver monitoring in simulators.
• Findings contribute to understanding human factors in mixed-autonomy driving environments.