A Predictive Model of a Driver's Target Trajectory Based on Estimated Driving Behaviors.

Area of Science:

• Automotive Engineering
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Driver behavior inference is vital for Advanced Driver Assistance Systems (ADAS).
• Existing methods often focus on discrete behavior labels (e.g., lane keeping).
• Predicting future driver trajectories offers a more nuanced approach.

Purpose of the Study:

• To develop a deep learning model for predicting driver trajectories.
• To represent driver behavior as polynomial functions for trajectory prediction.
• To enhance the capabilities of ADAS through accurate behavior forecasting.

Main Methods:

• Collected data from nine volunteers using a Driving Simulator experiment.
• Developed a deep learning network to predict polynomial coefficients for trajectories.
• Generated future vehicle trajectories based on predicted coefficients.

Main Results:

• The deep learning model effectively predicted coefficients for polynomial trajectory functions.
• Generated trajectories accurately represented the driver's likely path in the near future.
• Analysis identified factors influencing prediction errors.

Conclusions:

• The proposed deep learning model is effective for predicting driver target trajectories.
• This approach advances the potential for sophisticated ADAS design.
• Trajectory prediction offers a valuable alternative to discrete behavior labeling.