Driver lane-changing behaviours have a significant impact on the safety and the capacity of the vehicle-based traffic system. Therefore, modeling lane-changing maneuvers has become an essential component of driving behaviour analysis. Among microscopic LC models, game theory-based lane-changing models highlight the interaction of drivers, which reveal a more realistic image of driving behaviours compared to other classic models. However, the potential of game theory to describe the human driver's lane-changing strategies is currently under-estimated. This paper aims to review the recent development of game-theoretic models that are classified according to their different methodologies and features. They are designed for both human-driven and autonomous vehicles, and we hope they can find applications in future AV industries.

Changing lanes is a commonly-used technique for drivers to either overtake slow-moving cars or enter/exit highway ramps. Optional lane changes may save drivers travel time but increase the risk of collision with others. Drivers make such decisions based on experience and emotion rather than analysis, and thus may fail to select the best solution while in a dynamic state of flux. Unlike human drivers, autonomous vehicles can systematically analyze their surroundings and make real-time decisions accordingly. This paper develops a game theory-based lane-changing model by comparing two types of optimization methods. To realize our expectations, we need to first investigate the payoff function of drivers in discretionary lane-changing maneuvers and then quantify it in an equation of costs that trades-off safety and time-saving. After the evaluation for each alternative strategy combination, the results show that there exists a social gap in the discretionary lane-changing game. To deal with that problem, we provide some suggestions for future policy as well as autonomous vehicle controller designs, offering solutions to reduce the impact of disturbances and crashes caused by inappropriate lane changes, and also, inspire further research about more complex cases.

The emergence of vehicle-to-everything (V2X) technology offers new insights into intersection management. This, however, has also presented new challenges, such as the need to understand and model the interactions of traffic participants, including their competition and cooperation behaviors. Game theory has been widely adopted to study rationally selfish or cooperative behaviors during interactions and has been applied to advanced intersection management. In this paper, we review the application of game theory to intersection management and sort out relevant studies under various levels of intelligence and connectivity. First, the problem of urban intersection management and its challenges are briefly introduced. The basic elements of game theory specifically for intersection applications are then summarized. Next, we present the game-theoretic models and solutions that have been applied to intersection management. Finally, the limitations and potential opportunities for subsequent studies within the game-theoretic application to intersection management are discussed.

Machine learning algorithms aim to improve the power of predictors over conventional regression models. This study aims to tap the predictive potential of crash mechanismrelated variables using ensemble machine learning models. The results demonstrate selected models can predict severity at a high level of accuracy. The stacking model with a linear blender is preferred for the designed ensemble combination. Most bagging, boosting, and stacking algorithms perform well, indicating ensemble models are capable of improving upon individual models.