Lane-Changing Decision-Making Method for Autonomous Vehicles Considering Multi-Vehicle Interactions

Lane-changing decision-making is a critical element for ensuring the driving safety in complex driving environments for autonomous vehicles. In such environments, the interactions among multiple vehicles can significantly impact the safety of lane-changing decisions. To mitigate these risks, this study proposes a lane-changing decision-making method that takes into account the complex interactions between multiple vehicles. The proposed method integrates the long short-term memory (LSTM) network and the graph neural network (GNN) methods to make lane-changing decision-making in complex driving environments. Specifically, the LSTM network is utilized to capture the temporal dependencies of vehicle trajectories, which encode the underlying driving behaviors of individual vehicles. Meanwhile, the GNN is employed to model the interactions among multiple vehicles, which encompasses not only the interactions between autonomous vehicles and other vehicles, but also those among other vehicles. By jointly considering both types of interactions, a multi-vehicle interaction network is constructed based on the learned underlying driving behaviors of individual vehicles, which in turn is utilized to generate lane-changing decisions. The proposed method is verified with the dataset. Compared with the LSTM method, the accuracy of lane-changing decisions of the proposed method improved by 19.8%.