Trustworthy and Explainable AI for Traffic Signal Control: Concepts, Taxonomy, Challenges, and Research Directions

Traffic signal control (TSC) is a core component of intelligent transportation systems, directly influencing urban mobility efficiency, safety, and environmental sustainability. In recent years, deep reinforcement learning (DRL) and multi-agent reinforcement learning (MARL) have emerged as powerful paradigms for adaptive traffic signal control, consistently outperforming traditional fixed-time, actuated, and rule-based approaches. Despite these performance gains, the opacity of learning-based controllers remains a significant barrier to real-world deployment, particularly in safety-critical urban infrastructure where transparency, accountability, and regulatory compliance are essential. This paper presents a comprehensive survey of explainable artificial intelligence (XAI) for AI-based traffic signal control, with a particular focus on multi-agent systems. We systematically review state-of-the-art DRL- and MARL-based control methods and analyze their trustworthiness limitations, including non-stationarity, partial observability, emergent coordination behavior, and limited generalization. Building on this analysis, we provide a structured taxonomy of explainability techniques for traffic signal control, encompassing intrinsic interpretability, post-hoc explanations, reinforcement learning–specific explanations, and multi-agent interaction analysis. The survey further examines evaluation methodologies for both traffic performance and explanation quality, highlighting open challenges such as real-time explainability, scalability in large networks, and the trade-off between performance and interpretability. By synthesizing existing research and identifying key gaps, this work offers a consolidated reference for researchers and practitioners, and outlines future directions toward trustworthy and explainable AI-driven traffic signal control systems.

کلیدواژه ها:

Traffic Signal Control ، Deep Reinforcement Learning ، Multi-Agent Systems ، Explainable AI ، Trustworthy AI ، Survey