Modeling driver's evasive behavior during safety-critical lane changes: Two-dimensional time-to-collision and deep...

Area of Science:

• Traffic Safety Engineering
• Artificial Intelligence in Transportation
• Connected Vehicle Technology

Background:

• Lane changes are complex maneuvers often leading to safety-critical events.
• Existing traffic simulations and collision avoidance systems require improved models for evasive behaviors.

Purpose of the Study:

• To develop a lane-change-specific evasive behavior model.
• To enhance safety-aware traffic simulations and predictive collision avoidance systems.
• To utilize connected vehicle data for realistic behavior modeling.

Main Methods:

• Proposed a new surrogate safety measure: two-dimensional time-to-collision (2D-TTC) to identify critical situations.
• Utilized large-scale connected vehicle data from the Safety Pilot Model Deployment (SPMD) program.
• Applied a deep deterministic policy gradient (DDPG) algorithm for sequential decision-making modeling.

Main Results:

• Validated 2D-TTC by correlating detected risks with actual crashes.
• The DDPG model successfully replicated both longitudinal and lateral evasive maneuvers.
• Demonstrated the model's superiority in capturing complex driving behaviors during lane changes.

Conclusions:

• The developed evasive behavior model effectively simulates critical lane-change scenarios.
• The proposed 2D-TTC measure is a valid indicator of safety-critical driving events.
• This research advances the development of safer intelligent transportation systems.