Large real-world driving datasets have sparked significant research into various aspects of learning-based motion planners for autonomous driving. These include data augmentation, model architecture, reward design, training strategies, and planner pipelines. In this paper, we review and benchmark previous methods. Experiments show that many of these approaches have limited generalization abilities in planning performance due to overly complex designs or training paradigms. Experiments further reveal that as models are appropriately scaled, many designs become redundant. Therefore, we introduce StateTransformer-2 (STR2), a scalable, decoder-only motion planner. STR2uses a Vision Transformer (ViT) encoder and a mix-of-experts (MoE) causal transformer architecture. The MoE backbone addresses modality collapse and reward balancing by expert routing during training. Extensive experiments on the NuPlan dataset show that our method generalizes better than previous approaches across different test sets and closed-loop simulations. We evaluate its scalability on billions of real-world urban driving scenarios, demonstrating consistent accuracy improvements as both data and model size grow.

Ramp merging represents a bottleneck scenario that causes traffic congestion, accidents, and increases emissions. Connected and Automated Vehicles (CAVs) can realize the coordinated control of ramp merging through vehicle-to- infrastructure (V2I) for relieving above problems. Considering the previous studies on centralized ramp merging only involved single mainline, this paper proposes a multi-lane centralized collaborative control strategy using cooperative game. First, the merging rules of different lanes of vehicles in the merging area are defined, so that the vehicles can achieve the cooperative merging safely. Second, driving efficiency, comfort, and fuel consumption in the merging control zone are used as the cost function. The best merging sequences of vehicles in different lanes are solved by cooperative game. Finally, analytical solution of longitudinal optimal control for all vehicles is obtained by applying the Pontryagin principle. The effectiveness of the proposed method is verified through simulation under random traffic conditions. Compared to other centralized control algorithms, it can significantly improve driving efficiency and reduce fuel consumption. At the same time, the applicability of the method is verified by comparing with ExiD datasets, and some advantages are obtained in terms of fuel consumption.