Two-dimensional lane configuration design approach for Autonomous Vehicle Dedicated Lanes in urban networks

This study focuses on optimizing Autonomous Vehicle Dedicated Lanes (AVDLs) in urban networks, a critical step in managing mixed traffic where autonomous vehicles (AVs) and human-driven vehicles (HVs) coexist. Traditional AVDL deployment strategies have mainly optimized the number of AVDLs without adequately considering the directional functionality of lanes or their various lane-specific travel costs at intersections. To address these gaps, we propose a two-dimensional lane configuration approach that optimizes both the number of AVDLs on each road segment and their directional functionality for various traffic movements. Intersection delays are incorporated into the travel cost computation, through identifying the specific right-of-way allocations associated with different lane types. The proposed approach enables a more precise calculation of traffic volumes and travel costs on each lane-specific path, by categorizing travel into AVDL-only paths, hybrid-lane paths, and regular lane (RL)-only paths. A lane-specific user equilibrium (UE) model is developed to capture traffic dynamics on various lane types, with the existence and uniqueness of the UE solution rigorously proven. The AVDL configuration optimization is efficiently solved using a bi-level solution method. This method integrates a customized Monte Carlo Tree Search (MCTS) algorithm with a traffic accommodation ranking approach and a Frank–Wolfe-type algorithm with a link pruning technique to enhance computational efficiency. Numerical experiments on a toy network and the well-known Sioux-Falls network demonstrate the effectiveness of the proposed two-dimensional AVDL configuration approach and the efficiency of the bi-level solution method. This study contributes to the field by extending AVDL configuration to two dimensions, providing a comprehensive framework for future urban network design in mixed traffic environments.