Although prior studies have demonstrated the potential of connected and automated vehicle–dedicated lanes (CAV-DLs) to enhance traffic efficiency, their corridor-level impacts under mixed traffic environments have not been sufficiently quantified. A primary challenge in existing literature is the restriction of CAVs to dedicated lanes, a constraint that frequently triggers saturation imbalance by overloading CAV-DLs while leaving adjacent mixed lanes underutilized. To address these issues, this study develops a hierarchical control framework for mixed traffic arterials that integrates CAV lane selection with coordinated signal control. This framework utilizes a lane-selection mechanism to dynamically reallocate through-moving CAVs between dedicated and mixed lanes, complemented by a platoon-control policy in no-lane-change zones to exploit the benefits of reduced headways. Building on these mechanisms, a lane-selection-based multiagent proximal policy optimization (LS-MAPPO) controller is developed using the centralized training and decentralized execution (CTDE) paradigm. Extensive simulation results show that under high traffic demand, the LS-MAPPO controller reduces average vehicle delay by 18.6% to 40.3% compared with benchmark methods while simultaneously shortening queue lengths. The analysis further indicates that these benefits are highly sensitive to both traffic demands and CAV penetration rates (PRs). Specifically, under heavy demand with a 40% to 60% CAV PR, a single CAV-DL can reduce delay by 13.8% to 21.1%. In contrast, at low penetration levels, the deployment of CAV-DLs can be counterproductive, increasing delays by 31.7% to nearly 85.9% in extreme scenarios. These results provide quantitative evidence for the deployment of CAV-DLs across varying traffic demands and CAV PRs.
Chen et al. (Sat,) studied this question.