Multi-Objective reinforcement learning approach for improving safety at intersections with adaptive traffic signal...

Area of Science:

• Traffic Engineering
• Artificial Intelligence
• Transportation Safety

Background:

• Adaptive traffic signal control (ATSC) systems enhance traffic flow but have variable safety impacts.
• Proactive traffic safety improvement requires integrated efficiency and safety optimization.

Purpose of the Study:

• To develop and evaluate a safety-oriented ATSC algorithm for simultaneous optimization of traffic efficiency and safety.
• To introduce a novel multi-objective deep reinforcement learning framework for traffic signal control.

Main Methods:

• Utilized a multi-objective deep reinforcement learning framework.
• Trained and evaluated the algorithm on a simulated isolated intersection using real-world traffic data.
• Integrated a real-time crash prediction model for safety assessment.

Main Results:

• The proposed ATSC algorithm demonstrated improvements in both traffic efficiency and safety compared to benchmarks.
• Analysis revealed that abstracted control rules from the ATSC can enhance traditional signal controllers' safety performance.
• A hybrid controller was proposed for further safety enhancements.

Conclusions:

• The developed safety-oriented ATSC is the first to successfully optimize traffic safety.
• The findings suggest potential for improving traffic safety even with existing infrastructure through abstracted control rules.
• The research offers a pathway for more proactive traffic safety management in intelligent transportation systems.