The inclusion of unmanned aerial vehicles (UAVs) in conventional vehicular networks (VANET) facilitates efficacious and flexible data dissemination by enabling optimised solutions for disrupted communication while the physical obstacles limit the onboard units (OBUs)/vehicles(contact. Though a UAV-supported IoV network achieves an intelligent and promising solution for efficient data transfer in an infrastructure-less IoV network, malicious UAVs provide a huge penetration surface for attackers due to the unsatisfactory trustworthiness of the drones, which also leads to endangering the security and confidentiality of the users’ credentials. A robust authentication guarantees a potential solution for providing trust, credibility, and integrity. However, state-of-the-art authentication protocols in VANET suffer from ever-increasing cyber threats or huge computation overheads. In this paper, we have developed a blockchain-assisted, robust and efficient authentication and key agreement protocol for the VANET framework using Elliptic Curve Cryptography (ECC) where the interplanetary file system (IPFS) is employed to preserve the information blocks in an entirely distributed approach. The proposed protocol is formally appraised using a well-accepted Random Oracle Model (ROR) and simulated using a widely known AVISPA tool. We have also compared the performance of our scheme in terms of computation and communication overheads. All the results demonstrate the superior accomplishment and higher quality of service of the proposed scheme compared to other benchmark authentication schemes. This security-efficiency trade-off of our scheme makes it significantly adaptable in the real-time mobility environment.
Vehicular ad hoc networks (VANETs) are a crucial component of intelligent transportation systems; however, routing in V2V-based VANETs remains challenging due to dynamic topologies, incomplete observations, and the limited resources of edge devices. Existing reinforcement learning approaches often assume fixed graph structures or require retraining when network conditions change, making them unsuitable for deployment on constrained hardware. This paper presents TrajAware, an RL-based framework designed for edge AI deployment in VANETs. TrajAware integrates three components: (i) trajectory-aware prediction, which uses historical routes and junction information to estimate real-time positions under incomplete observations; (ii) action space pruning, which reduces redundant neighbour options while preserving two-hop reachability, alleviating the curse of dimensionality; and (iii) a permutation-equivariant set encoder, which processes neighbour sets while preserving neighbour-representation alignment under input permutations. TrajAware is evaluated in the open-source SUMO simulator using real-world city maps with a leave-one-city-out setup. Results show that TrajAware achieves near-shortest paths and high delivery ratios while maintaining efficiency suitable for constrained edge devices, outperforming state-of-the-art baselines in both full and incomplete observation scenarios.
Autonomous driving systems face considerable challenges in maintaining safety and human-like behavior within uncertain and partially observable environments. To address these challenges, we propose UDiffSAC-HL, a perceptual uncertainty-aware Diffusion Soft Actor-Critic (SAC) framework with a Human-in-the-Loop (HITL) component, aimed at enhancing decision-making robustness and safety in complex driving scenarios. The proposed framework features a HITL mechanism with novel dual-level intervention and a rule-based safety controller, ensuring minimal safety guarantees even in extreme conditions. Within this framework, the conventional actor network of SAC is replaced by a diffusion-based policy generator, which enables stable and interpretable action generation with a straightforward reward design. To tackle perception uncertainty caused by occlusions, we propose a novel Diffusion Uncertainty-aware Sampling Model (DUSM) that integrates uncertainty-guided denoising and human preference alignment in both the training and sampling objectives. Specifically, a structured uncertainty modeling technique is developed to jointly encode the field-of-view loss ratio and the trajectories of hidden potential agents, thereby quantifying and incorporating perceptual uncertainty into the policy generation process. Experimental results in the CARLA simulator demonstrate that UDiffSAC-HL outperforms benchmark methods in terms of policy learning performance, efficiency, success rate, and collision avoidance, while also producing more human-like, uncertainty-aware driving behaviors.
As global ports accelerate their automation transition, uncertain disruptions have become a critical barrier to sustaining operational efficiency in automated container terminals (ACTs). Complex unplanned priority tasks (CUPT) may involve diverse scales, varying priorities, and overlapping time windows. When multiple batches of CUPT are urgently inserted into planned loading operations, they can easily disrupt a terminal operating system (TOS) that has limited resilience, leading to scheduling disorder. Existing studies often model task attributes via (non-)homogeneous Poisson processes. However, their independence assumption can be restrictive when CUPT generate clustered arrivals, sequential dependence, and cross-task interactions. To address this gap, this study models loading operations under CUPT disruptions using a multivariate Hawkes process, which captures both self-exciting intra-task dependence and mutually exciting inter-task interaction. Based on this model, we propose a rule-heuristic multi-agent deep deterministic policy gradient (RH-MADDPG) framework. It integrates global real-time information and mathematically grounded heuristic rules into a multi-agent system to achieve the optimization objectives implied by the multivariate Hawkes framework. Extensive simulation experiments based on operational parameters from the Tianjin Port ACT, including a Hawkes-mechanism consistency check, progressive algorithm comparison, sensitivity analysis, robustness tests under different CUPT arrival scenarios, and ablation studies, demonstrate that RH-MADDPG effectively prioritizes CUPT while maintaining the efficiency of regular task, consistently outperforming benchmark methods across key evaluation indicators. Its advantage also remains under a vertical yard layout. The findings provide actionable insights for port operators, shipping lines, and cargo owners, enabling a proactive and resilient response to CUPT-related disruptions.
