Analysis of AV merging behavior in mixed traffic using large-scale AV driving datasets

Area of Science:

• Traffic Engineering
• Autonomous Systems
• Behavioral Analysis

Background:

• Autonomous vehicles (AVs) offer potential advantages in complex maneuvers like merging compared to human-driven vehicles (HDVs).
• Limited research exists on AV-HDV interactions during real-world merging events, necessitating analysis of actual driving data.

Purpose of the Study:

• To analyze AV merging behavior in mixed traffic using real-world datasets.
• To investigate the impact of traffic variables on merging gap time (GT) and crash risk.
• To compare AV and HDV merging performance and safety.

Main Methods:

• Extraction and analysis of merging events from Argoverse-2 and Waymo AV testing datasets.
• Development of a Weibull random parameter hazard-based duration model to assess GT determinants.
• Application of extreme value theory (EVT) for merging crash risk estimation.

Main Results:

• AVs and HDVs exhibited similar merging gap time distributions.
• Increased merging gap time reduced speed variations in the target lane for both AVs and HDVs.
• Crash risk was comparable between AV-involved and HDV-only merging events, but AVs reduced crash severity.

Conclusions:

• AVs demonstrate potential for safer merging in mixed traffic due to programmed behaviors like smoother acceleration.
• Optimizing AV algorithms for dynamic, human-centric conditions is crucial for enhancing safety and efficiency in complex scenarios.
• Further AV technology improvements are needed for seamless integration into mixed traffic environments.