Reinforcement Learning (RL) has emerged as a powerful paradigm for adaptive traffic signal control. However, the opaque, black-box nature of deep RL models poses challenges in safety-critical infrastructures such as traffic control, affecting acceptance by transportation agencies, regulatory compliance, operational trust, troubleshooting, and fine-tuning. To bridge this gap between high-performance optimization and human-comprehensible interpretability, this work introduces a novel explainable entity-centric RL framework for safe and transparent traffic signal control.
Instead of processing traffic states through monolithic, flat vectors, the proposed architecture disaggregates real-time intersection observations into distinct, high-dimensional lane entities and phase temporal configurations, preserving the structural topology and geometric configurations of the intersection. Relational dependencies and inter-lane conflicts are dynamically extracted via a dual-stage attention network featuring sequential multi-head cross-attention and self-attention blocks. This design produces a real-time affinity matrix that quantifies the direct influence of signal phases on specific approach volumes and queues, providing full visual and analytical interpretability.
To ensure strict operational reliability, a deterministic action-masking interface is integrated directly into the Proximal Policy Optimization (PPO) pipeline, explicitly blocking invalid phase transitions to guarantee absolute compliance with established signal timing and safety constraints. Evaluated in a microscopic simulation environment, it outperforms state-of-the-art baselines in delay minimization. More importantly, the emergent attention weights align precisely with established traffic engineering principles, offering an auditable, trust-enabling, and deployable architecture for next-generation adaptive traffic control systems.
Blogger's Review: This paper introduces an innovative explainable reinforcement learning framework that optimizes traffic signal control while enhancing system transparency and reliability. The emphasis on interpretability is crucial in safety-critical domains, ensuring compliance and public trust in traffic management. This research not only holds theoretical significance but also demonstrates strong potential for practical applications.