Impact of traffic management center functionality on speed harmonization in urban networks under mixed traffic environments

The effectiveness of speed harmonization using connected and autonomous vehicles (CAVs) in urban mixed traffic depends on both traffic management center (TMC) functionality and the data available for control. Using multi-agent reinforcement learning (MARL), this study examines whether performance differences among TMC functionality configurations vary with data availability. Three TMC functionality configurations are considered, represented by distinct training and execution paradigms: centralized training and execution (CTE), centralized training with decentralized execution (CTDE), and decentralized training and execution (DTE). Each configuration is evaluated under standalone autonomous vehicle (SAV), infrastructure-assisted (IA), and fully connected (FC) conditions, representing progressively broader access to onboard sensing, infrastructure data, and planned route information. Experiments are conducted with 20 and 40 simultaneously controlled CAVs in a calibrated microsimulation and supplemented by an FC experiment in the TAVF-Hamburg network, where the existing sensing and communication infrastructure corresponds to the FC condition. CTE achieves the most consistent performance across SAV, IA, and FC, with the strongest overall results under FC. Under SAV and IA conditions with 20 CAVs, CTDE shows pronounced performance degradation, whereas its performance improves substantially with 40 CAVs. The independent-learner DTE configuration exhibits considerable variability in performance across data availability and fleet-size conditions. The results reveal an asymmetric and configuration-dependent relationship between TMC coordination authority and data availability. These findings indicate that institutional constraints on TMC functionality and regulatory restrictions on data access and sharing should be considered jointly when selecting configurations for early CAV deployment.