LFF: An attention allocation-based following behavior framework in lane-free environments

With the rapid advancement of autonomous driving technology, current autonomous vehicles (AVs) typically rely on lane markings and parameters for operation despite their advanced perception capabilities. This research aims to develop a Lane-Free Following (LFF) framework to address behavior planning for AVs in environments lacking clear lane markings. The LFF utilizes decision modules, such as Monitoring Zones, Focus Zones, and Passing Corridors, to dynamically select the most appropriate following strategy. It integrates a Multi-Target Following Model (MTIDM) and an attention allocation mechanism to optimize acceleration control by adjusting attention concentration levels. Initially, we examine the stability of multi-target following and determine the stability region on a two-dimensional plane using specific stability criteria. Subsequently, the LFF is integrated with the lateral model of the Intelligent Agent Model (IAM), and calibrated and validated using lane-free traffic data from Hefei, China, and Chennai, India. Simulation results demonstrate the LFF's high accuracy across various vehicle types. In simulations conducted on open boundary roads and virtual circular roads with varying widths and traffic densities, the LFF showed enhanced driving comfort and efficiency. This optimization of road widths and densities improved traffic flow and road space utilization compared to traditional lane-based traffic. In congested start conditions on circular roads, we compared the uniform attention allocation mode (LFF-UA), the concentrated attention allocation mode (LFF-CA), and the High-Speed Social Force Model (HSFM). Results indicated that the HSFM excels in velocity and flow, offering faster startup efficiency. The LFF-UA, while maintaining efficiency, evenly distributed attention to neighboring preceding vehicles, enhancing driving safety and reducing fuel consumption and emissions. This research addresses current issues in mixed traffic environments and provides theoretical references for the future application of connected autonomous vehicles in lane-free environments.