Modeling Lane-Changing Behaviors for Autonomous Vehicles Based on Vehicle-to-Vehicle Communication

With the advent of autonomous vehicles (AVs) and advanced driving assistance systems (ADAS), there has been a growing interest in studying driving behaviors within the field of transportation science. Given that the transition period of mixed traffic is expected to continue for more than 30 years, it is crucial to evolve AV technology to resemble human driving, especially in the freeway weaving sections. Lane-changing (LC) maneuvers in these sections could cause problems for traffic flow, such as traffic breakdown, oscillation, or bottleneck activation. This study proposes an interpretable LC implementation model for naturalistic driving behaviors of AVs based on vehicle-to-vehicle (V2V) communication. To achieve this objective, a systematic selection process is adopted to find optimal V2V features that resemble how human drivers assess LC situations. Based on the minimum redundancy maximum relevance (mRMR) algorithm, seven V2V features have been selected out of 25 candidates. Then, a support vector machine (SVM) is employed to investigate how these features exhibit in each of LC and lane-keeping (LK) situations. The proposed model was applied in a field case of a weaving Section on freeway US 101. Performance measures of simple accuracy, precision, recall, and F1-score show high accuracy of 0.9814, 0.9150, 0.7955, and 0.8511, respectively. Subsequently, a strategy for naturalistic LC behaviors of AVs was simulated. The proposed model outperforms high prediction accuracy compared to other existing models. Particularly, errors in the lateral movements have significantly improved. These results suggest that the proposed model effectively simulates naturalistic LC behaviors based on V2V communication.