A Study of Data-Driven Free Lane Changing Strategy on the Highway

Autonomous driving is a key topic in the automotive industry, and lane-changing research is a crucial component of it. This paper, based on existing studies, presents a data-driven humanoid lane-change decision model. First, a vehicle lane-changing dataset is constructed using the HighD dataset as the original data source. The dataset focuses on filtering vehicle trajectories for free lane changes in highway scenarios, applying Kalman filtering for noise reduction, and then analyzing feature variables using the Maximal Information Coefficient (MIC). Subsequently, a model of free lane-changing strategy is constructed for highway scenarios based on machine learning. The lane-changing decision models, based on decision trees and support vector machines (SVM), are trained, with model parameters optimized using Bayesian optimization algorithms. Finally, a performance comparison is conducted between the two lane-changing decision models. The results indicate that the SVM-based lane-changing decision model outperforms the decision tree model.