Conflict resolution behavior of autonomous vehicles at intersections under mixed traffic environment

Area of Science:

• Traffic Engineering
• Autonomous Systems
• Road Safety

Background:

• Intersection navigation is complex for autonomous vehicles (AVs) due to varied road user types and traffic dynamics.
• Understanding AV driving behavior at intersections is crucial for safe integration into mixed traffic environments.

Purpose of the Study:

• To investigate intersection-related traffic conflicts involving AVs.
• To analyze AV driving behavior and its impact on relative crash risks using the Arogoverse-2 dataset.

Main Methods:

• Categorized conflict scenarios into AV-involved (AV-first, AV-second) and no AV conflict.
• Applied agglomerative hierarchical clustering with t-SNE for driving style categorization.
• Utilized a three-layer Bayesian hierarchical model to assess driving volatility and traffic characteristics' effect on crash risks.

Main Results:

• 29% of AV-first conflict events posed high risks, while AV-second events were low or medium risk.
• AVs demonstrated safer interactions with human-driven vehicles, pedestrians, and cyclists, especially at higher, uniform speeds.
• AVs exhibited safer behavior during unprotected left turns and with vulnerable road users compared to human-driven vehicles.

Conclusions:

• AVs generally exhibit safer driving behaviors at intersections, particularly when yielding and interacting with vulnerable road users.
• Findings highlight areas for AV technology improvement to enhance adaptation to diverse intersection types and mixed traffic conditions.