From ambiguity to precision: Digitized traffic rules for autonomous driving

Area of Science:

• Intelligent Transportation Systems
• Robotics and Control Systems
• Traffic Engineering

Background:

• Human-driven and automated vehicles (AVs) interact in mixed traffic, necessitating AVs to understand and follow human driving rules.
• Existing traffic rules often use ambiguous language (e.g., 'not impede'), posing challenges for AV implementation and safety.
• Accurate inference and compliance with traffic rules are critical for AV safety in complex driving scenarios.

Purpose of the Study:

• To develop and validate a rule optimization framework for AVs that bridges the gap between ambiguous natural language traffic rules and precise AV implementation.
• To enhance the safety, operational efficiency, and ride comfort of AVs in mixed traffic environments.
• To ensure AV behavior aligns with human driving norms and traffic regulations.

Main Methods:

• A hybrid approach combining knowledge-based reasoning and data-driven optimization using safety-critical event (SCE) data from real-world intersections.
• Semantic classification of traffic rules, formalization using Metric Temporal Logic (MTL), and parameter supplementation from regulatory documents.
• Calibration of optimal parameters using a genetic algorithm within simulation environments reconstructed from SCE data.

Main Results:

• The rule-integrated planner demonstrated a safety performance enhancement of over 90% compared to human drivers, while maintaining operational efficiency and ride comfort.
• Optimized rules integrated into existing planners on CommonRoad and INTERACTION benchmarks consistently reduced collision risks, showing strong generalizability.
• Key performance drivers identified as parameters controlling turn ranges, temporal safety thresholds, and longitudinal distances.

Conclusions:

• The proposed framework offers an interpretable and transferable solution for optimizing traffic rules, enhancing AV safety and alignment with human driving.
• This research addresses the challenge of implementing ambiguous traffic rules in AVs, paving the way for safer mixed-traffic environments.
• The findings highlight the importance of parameter calibration for achieving robust and safe AV behavior in diverse driving conditions.