Abstract Highway ramp merging areas are critical bottlenecks affecting traffic safety and operational efficiency, where interactions between mainline and ramp vehicles often induce traffic conflicts, crashes, and delays. Connected and automated vehicles (CAVs), enabled by precise trajectory control and real-time cooperation, provide a promising solution for improving ramp merging operations. Existing reviews mainly compare algorithms or application scenarios, with limited attention to the physical process of ramp merging, in which longitudinal speed and spacing adjustments must be coordinated with lateral lane-changing motion to enter a safe mainline gap within a limited merging area. To examine how this motion is planned and realized, this paper develops a decision-planning and control-execution framework for reviewing CAV ramp merging strategies. The two functional modules are analyzed across individual-vehicle, general multi-vehicle, and platoon-based forms of vehicle organization, focusing on their technical methods, implementation, and effects on traffic performance. Their system-level organization is then compared under hierarchical cooperative, integrated, and hybrid control architectures. Finally, future research directions for CAV ramp merging control are identified.
Abstract The widely adopted SAE (Society of Automotive Engineers) taxonomy categorizes driving automation primarily based on the division of operational responsibilities between the human driver and the system. This has worked well for several applications. However, it fails to distinguish the requisite versatility of autonomous vehicle (AV) performance across diverse operational environments. This structural limitation has spawned critical challenges, including safety risks, commercialization barriers, regulation challenges, and inadequate Operational Design Domain (ODD) development guidelines toward Level 5 autonomy. To address these gaps, this paper proposes a supplemental taxonomy for SAE Levels 3 and 4 using an Environmental-Situational Capability (ESC) evaluation framework across three core dimensions: the built environment, natural environment, and ambient traffic conditions. Using a two-stage hierarchical classification strategy, the ESC framework maps expert evaluations of a specific automated driving system (ADS) to corresponding levels of the proposed taxonomy. This novel classification bridges the ADS level and the ODD by correlating environmental complexity with the minimum capabilities required for safe autonomous operation. Operating orthogonally to SAE standards, this taxonomy categorizes systems based on their actual operational capabilities rather than driving automation task responsibility alone. Consequently, it establishes a crucial alignment across researchers, manufacturers, regulators, and consumers, providing a practical reference for product benchmarking, safety validation, and policymaking.
Abstract In autonomous driving, the continuous movement of objects exacerbates point cloud sparsity, making future frame prediction a challenging task. While self-supervised methods excel in such dynamic environments by learning temporal evolution directly from unannotated data, existing approaches often model spatial structures and temporal dynamics independently, limiting their effectiveness in sparse and complex scenes. To overcome this limitation, we reformulate point cloud prediction as a unified spatio-temporal correlation learning problem and propose a novel framework, the Motion-guided Spatio-Temporal Point Cloud Prediction Network (MoCNet). The key idea is to use motion information to guide feature learning. Specifically, we design a dual-branch structure to separately capture spatial structure and temporal variation, and then integrate them into a unified representation. To better model long-range dependencies, we adopt a large-kernel convolution strategy to enhance feature interactions without introducing heavy computational cost. In addition, we employ a dual-frequency representation scheme that decomposes features into high-frequency and low-frequency components. This design helps preserve local details while maintaining global consistency. Experimental results show that the proposed method consistently outperforms existing approaches on multiple benchmark datasets. It achieves higher prediction accuracy and better generalization ability. Moreover, MoCNet supports real-time inference, with an average latency of 20 ms and 15 ms under different settings. This is significantly faster than the typical 3D LiDAR sampling rate. The predicted point clouds are also closer to ground truth in terms of shape, orientation, and spatial position, which improves reliability for downstream autonomous driving tasks.
Abstract This paper proposes a reinforcement learning (RL) based optimal platoon control framework for autonomous vehicles (AVs) subject to uncertainties and cyberattacks. Within the optimized backstepping design, an actor-critic-identifier architecture is employed to improve platoon control performance. A cyberattack compensation mechanism and a fuzzy logic based composite observer are integrated to reconstruct reliable vehicle information and estimate system uncertainties under compromised measurements, including the internal vehicular system states, uncertainties, and cyberattacks. In addition, a second-order tracking differentiator is introduced to smooth transient responses and substantially reduce the computational burden associated with differentiating the virtual controllers generated during the backstepping procedure. Based on the Lyapunov function based theory, the stability of the entire vehicle platoon is established, and all relevant errors in the multi-vehicle system are proven to be semiglobally uniformly ultimately bounded (SGUUB). Finally, numerical simulations of a multi-vehicle platooning system validate the effectiveness of the proposed framework and demonstrate its robustness against the simultaneous presence of uncertainties and cyberattacks. Moreover, a comparative study with a non-optimal control algorithm is conducted, highlighting the superior tracking performance and transient response of the proposed RL-enabled optimal control framework.
As large language models (LLMs) are increasingly integrated into autonomous vehicles (AVs), understanding their reasoning and behavioral tendencies becomes essential. Trained on vast datasets, LLMs carry behavioral priors and social biases that may shape their driving decisions. Without such insight, developers may struggle to align models for AV needs. To address this, we probe prompt-conditioned high-level action choices of LLMs, each with 1500 contextual variants. Three widely used LLMs are evaluated with multilingual prompts to select from predefined behavioral options ordered by aggressiveness. An ordered logit model quantifies how contextual factors influence decisions, complemented by thematic analysis to reveal underlying reasoning tendencies. The results show that LLM decisions reflect a mix of model characteristics, linguistic framing, and scenario context. Across conditions, models remain sensitive to rider urgency, traffic complexity, and road-user types. Generative Pre-trained Transformer(GPT)is more conservative, while DeepSeek and LLaMA act more assertively, especially in vehicle interactions. Prompt language also matters. Chinese and French prompts are associated with more assertive behavior than English prompts, with French being the strongest. Across scenarios, all models shift toward more protective behavior when vulnerable road users appear, reducing aggressiveness and prioritizing safety and smoother flow. These findings characterize model-level behavioral priors relevant to LLM choice and prompt design in AV applications.
