Iterative learning control for lane-changing trajectories upstream off-ramp bottlenecks and safety evaluation

Area of Science:

• Intelligent Transportation Systems
• Control Theory
• Machine Learning

Background:

• Highway off-ramp diverging areas present complex traffic challenges, particularly for connected and automated vehicles (CAVs).
• Existing traffic control methods struggle to optimize lane-changing maneuvers in these high-risk zones.
• Improving traffic operation and safety requires advanced control strategies for CAVs.

Purpose of the Study:

• To propose and evaluate an iterative learning control (ILC) framework for CAV lane changing.
• To enhance traffic operation and safety in highway off-ramp bottleneck areas.
• To reduce both longitudinal and lateral collision risks through optimized trajectory imitation.

Main Methods:

• Developed an iterative learning control framework for CAVs at highway off-ramps.
• Utilized the Next Generation Simulation (NGSIM) dataset, filtered by a cost function.
• Employed Random Forest (RF) and Back Propagation Neural Network (BPNN) for lane-changing decision (LCD) and execution (LCE) models.
• Incorporated an iterative learning approach where successful simulation data refines the models.

Main Results:

• The iterative framework reduced longitudinal risk by 50% and lateral risk by 28.7%.
• At 100% CAV market penetration rate (MPR), longitudinal and lateral risks decreased by 90% and 35%, respectively.
• Significant system value-at-risk improvements were observed once CAV MPR reached 50%.

Conclusions:

• The proposed iterative learning control framework effectively improves safety and traffic flow for CAVs in complex off-ramp scenarios.
• The framework demonstrates substantial risk reduction capabilities, particularly at higher CAV market penetration rates.
• This approach offers a promising solution for managing CAVs in critical highway infrastructure.