Making a few talk for the many - Modeling driver behavior using synthetic populations generated from experimental...

Area of Science:

• Automotive Engineering
• Human Factors in Transportation
• Computational Statistics

Background:

• Advanced driver assistance systems (ADAS) rely on accurate driver behavior models.
• Experimental studies are crucial for model development but face data collection cost barriers.
• Limited participant numbers in experiments restrict the generalizability of driver models.

Purpose of the Study:

• To develop a methodology for improving the reliability of experimental driver studies with limited participants.
• To create a virtual population of drivers using Bayesian inference to generate synthetic data.
• To enhance the scope of experimental data by simulating realistic variations in driver behavior.

Main Methods:

• Developed a Bayesian inference framework to generate synthetic driving cases.
• Ensured synthetic data adheres to real-world constraints and observed experimental variations.
• Applied the methodology to truck driver right-turn maneuver data, including scenarios with crossing cyclists.

Main Results:

• The methodology successfully generated synthetic speed profiles during braking that mimicked observed data.
• Synthetic profiles extended to include realistic braking patterns not initially observed in the experimental data.
• The framework demonstrated the ability to create diverse yet realistic driving behaviors based on speed profiles and physical constraints.

Conclusions:

• The developed Bayesian inference methodology effectively creates virtual driver populations to augment experimental data.
• This approach enhances the reliability and scope of driver models derived from limited experimental studies.
• The framework shows significant promise for the automotive industry in designing active safety systems and automated driving technologies.