Online Personalized Preference Learning Method Based on In-Formative Query for Lane Centering Control Trajectory.

Area of Science:

• Human-Computer Interaction
• Robotics
• Artificial Intelligence

Background:

• Personalization of autonomous vehicles (AVs) and advanced driver assistance systems (ADAS) often assumes drivers want vehicles to imitate their own driving.
• This assumption may not be universally true, as individual driving preferences vary significantly.
• Existing methods lack a robust approach to learn these unique, personalized driving preferences.

Purpose of the Study:

• To propose an online personalized preference learning method (OPPLM) that accurately learns individual driver preferences for AVs.
• To address the limitation of current systems that assume a one-size-fits-all driving style.
• To develop a system that adapts to each driver's unique trajectory preferences.

Main Methods:

• Utilizes a pairwise comparison group preference query and a Bayesian approach for learning.
• Employs a two-layer hierarchical structure model based on utility theory to represent driver preferences.
• Models uncertainty in driver query responses and uses informative and greedy query selection for efficiency.
• Proposes a convergence criterion to determine when a driver's preferred trajectory is found.

Main Results:

• The OPPLM demonstrated rapid convergence, requiring an average of only 11 queries in a user study.
• Successfully learned the driver's preferred trajectory in the context of a lane centering control (LCC) system.
• The estimated utility from the driver preference model showed high consistency with subject evaluation scores.

Conclusions:

• The proposed OPPLM effectively and efficiently learns personalized driving preferences for autonomous vehicles.
• This method overcomes the limitations of driver-imitation approaches by adapting to individual driver needs.
• The findings support the development of more sophisticated and user-centric autonomous driving systems.