Toward Trustworthy XAI-Based Multi-Agent Traffic Signal Control with Human-in-the-Loop Oversight

Artificial intelligence–based traffic signal control has shown strong potential to improve urban traffic efficiency through adaptive and decentralized decision-making. In particular, deep reinforcement learning (DRL) and multi-agent reinforcement learning (MARL) enable real-time coordination across complex traffic networks. However, the black-box nature of such controllers undermines trust, limits human oversight, and hinders deployment in safety-critical urban environments. This paper proposes a conceptual framework for trustworthy explainable AI (XAI)-based traffic signal control that integrates multi-agent deep reinforcement learning with real-time explainability and human-in-the-loop (HITL) oversight. Explainability is embedded directly into the control loop, enabling multi-level explanations at the feature, action, policy, and network levels, as well as counterfactual reasoning to clarify decision boundaries and system behavior. A dedicated human oversight layer allows traffic engineers to monitor, validate, and intervene in AI-driven control decisions, supporting continuous learning while preserving safety and accountability. By explicitly co-optimizing traffic performance and interpretability, the proposed framework addresses key deployment challenges related to transparency, robustness, and fairness. Overall, this work provides a structured pathway toward auditable, human-aligned, and deployable AI-based traffic signal control for future smart cities.

کلیدواژه ها:

Explainable AI ، Multi-Agent Reinforcement Learning ، Traffic Signal Control ، Human-in-the-Loop ، Trustworthy AI