Safe and Efficient Traffic Control with Constrained Reinforcement Learning Models

Traffic congestion remains a significant challenge in modern urban planning, leading to increased travel times, fuel consumption, and greenhouse gas emissions. While traditional traffic management systems often fail to adapt to dynamic traffic patterns, Reinforcement Learning (RL) offers a promising solution by enabling adaptive and intelligent traffic control. However, ensuring safety and efficiency in RL-based traffic systems remains a critical challenge, especially when dealing with real-world constraints such as pedestrian safety, speed limits, and emergency vehicle prioritization. The study develops a safe RL framework that integrates domain-specific constraints using techniques such as reward shaping, Lagrangian relaxation, and constrained policy optimization.

کلیدواژه ها:

Traffic Control ، Reinforcement Learning ، Congestion Reduction