Dynamic lane management for emerging mixed traffic with semi-autonomous vehicles

Semi-autonomous vehicles (semi-AVs), situated between fully-autonomous vehicles (full-AVs) and traditional vehicles (TVs), offer functionalities that allow drivers to activate autonomous driving features. These functionalities, such as Tesla Autopilot, BMW Personal Pilot, and General Motors Super Cruise, relieve drivers at the wheel of certain driving tasks and are anticipated to enhance road capacity by improving driving efficiency. However, the immaturity of early- stage autonomous driving technology can hinder immediate improvements in road capacity, especially in scenarios with a mix traffic of manually and autonomously driven vehicles. To mitigate these challenges, this study introduces the use of dedicated lanes and designs an intelligent corridor system that dynamically optimizes the allocation of lanes for auto-driven and human-driven vehicles. Firstly, a congestion model is established to capture the dynamics of bottleneck congestion and derive vehicle delays with demand changes, serving as a valuable reference for developing lane management strategies. Then, the choice of driving mode for semi-AVs is bounded with lane selection and modeled using a dynamic user equilibrium model over discrete-time series. Based on that, numbers of auto-driven and human-driven lanes are dynamically optimized with the objective of minimizing system costs. To prevent frequent adjustments of lane types that could degrade system performance, we employ a non-myopic decision-making strategy to account for both immediate and future costs, ensuring robust and efficient lane management over the entire decision horizon. Through numerical experiments, we validate the effectiveness of the dynamic lane management and non-myopic strategy under various semi-AV penetration and demand levels, demonstrating that dynamic lane management outperforms (or at least equals) fixed-lane scenarios and static lane management in all test scenarios. Additionally, we conducted sensitivity analyses on AV adoption levels, demand levels, decision horizons, and period lengths, uncovering useful insights for the practical application of dynamic lane management. Overall, this study offers a promising solution to efficient lane management of corridor systems in mixed traffic, especially in the initial stage of AV adoption.