The development of intelligent transportation systems (ITS) has made advanced and comprehensive simulation essential for the safe and efficient evaluation of connected and autonomous vehicles (CAVs). This paper addresses the crucial role of co-simulation in advancing ITS and CAV technologies, highlighting the limitations of single-platform simulations in capturing the benefits of integrated simulation environments. We explore and classify various simulators, based on their tasks within autonomous driving systems (ADS), and compare open-source and commercial simulators. Through a set of case studies, we demonstrate how co-simulation enhances the realism and comprehensiveness of testing frameworks, addressing the multifaceted challenges in ITS. Despite technical challenges such as fidelity, data handling, integration, and human factors, co-simulation provides a robust approach to developing innovative and safer urban mobility solutions. Future advances in artificial intelligence (AI), machine learning (ML), and quantum computing may further enhance co-simulation capabilities, thereby fostering safer and more efficient transportation simulations for academia, industry, and regulatory bodies.

Urban traffic networks face increasing challenges in optimizing signal control under dynamic and heterogeneous conditions, while adhering to real-world operational require ments such as minimum green times, pedestrian recalls, and coor dination rules. Traditional centralized approaches often struggle to adapt in real time or generalize across diverse intersections. This paper presents a decentralized graph-based multi-agent reinforcement learning (MARL) framework for adaptive traffic signal control. Each intersection operates as an independent agent with directional state representations, while a multi-head Graph Attention Network (GAT) enables efficient neighbor communication for both localized and coordinated decisions. Real-world SPaT timing rules, including green durations, clearance intervals, and pedestrian service requirements, are enforced through rule-based feasibility checks within the environment rather than learned as soft constraints. A two-stage decision process first identifies the dominant traffic direction and then determines whether to maintain or switch phases under constraint-aware action masking. The reward function penalizes vehicle delay, stops, and excessive switching. Experiments conducted using a high-fidelity simulation of a real-world urban corridor demonstrate consistent efficiency gains across varying demand levels while preserving operational feasibility. Designed for IoT-enabled deployment, the framework supports real-time controller communication via NTCIP and edge modules, while quantitative analysis of communication bandwidth, latency, and asynchronous updates is left for future field-oriented evaluation.

1. Urban traffic flow is frequently disrupted by repeated stop-and-go movements and vehicle queues at intersections, resulting in significant delays and substantial traffic oscillations (Urbanik et...