Federated learning is a promising paradigm for achieving distributed intelligence by protecting user privacy in vehicular networks. Considering limited computing and communication resources, it is important to select appropriate clients from a huge number of users to participate in the training process. In vehicular networks, the problem of choosing proper clients is particularly complex due to the heterogeneity of network users, including the differences in the data, computation capability, available throughput, and samples freshness. We design a fuzzy logic based client selection scheme to address this issue. The proposed scheme considers the number of local samples, samples freshness, computation capability, and available network throughput based on a fuzzy logic approach. Extensive simulation results show that the proposed scheme outperforms other baselines.

Vehicle platoon is a group of vehicles moving in the same direction at a certain speed while maintaining a stable inter-vehicle distance, and so the vehicles are closely connected with each other in a virtual way. While a vehicle platoon can improve traffic efficiency, there are some challenges in maintaining communication among platoon members, especially when some vehicles join an existing platoon that includes long-body, heavy-duty or special vehicles. In order to solve this problem, we propose a spectrum sensing scheduling (SSS) scheme for communication resource management in platooning. First, we propose a three-level platoon architecture, which includes the cloud, road side unit (RSU), and vehicles to increase the communication coverage of a platoon and multiple platoons. Second, the spectrum sensing model, platoon control error model, and platoon communication delay model are established for the platoon delay under the SSS scheme. Third, we propose a greedy algorithm for resource allocation, which is based on the SSS scheme and vehicle-to-vehicle (V2V) communications, to minimize platoon delay. Finally, we simulate platoon delay and platoon safety status of the SSS scheme when some vehicles join an existing platoon. Simulation results show that the proposed SSS scheme reduces the platoon communication delay by around 50% and achieves a smaller platoon error as compared with existing baseline schemes for resource scheduling.

With the development of advanced information and communication technology, the traditional centralized cloud architecture cannot satisfy the exploding demand for data exchange in Internet of Vehicle (IoV) systems. Moreover, the traditional centralized architecture of the vehicular network has the potential risk of a single point of failure and lacks autonomy since the system highly relies on a trusted third party (TTP) to provide identity management. Fortunately, the emergence of blockchain technology provides a potential direction to address these problems. However, there are still some problems existing in the construction of an efficient blockchain system in IoV systems, such as the dynamic network topology and limited resources. In this paper, we propose a hierarchical resource scheduling scheme for blockchain-enabled IoV systems that improves the performance of the blockchain-enabled IoV system by efficiently allocating computational resources. The superiority of the proposed method is fully demonstrated by comparing it with existing baseline methods.

Autonomous driving technology has achieved remarkable advancements, offering substantial potential to revolutionize traffic safety and smart mobility. However, when faced with rare scenarios (weather, accident scenes, lighting), autonomous driving systems can still only play a limited role due to insufficient learning in these rare situations. To address this challenge, we propose a novel approach that leverages Low-Rank Adaptation (LoRA) and Mixture-of-Experts (MoEs) technologies to enhance the performance of pre-trained autonomous driving models in handling rare situations. Specifically, we firstly use LoRA to fine-tune the pre-trained model of autonomous driving to focus on capturing knowledge related to rare scenarios and enhance the model’s ability to handle rare situations. Furthermore, we introduce MoEs and propose local, global, and hybrid adaptive solutions to overcome LoRA’s fixed intrinsic rank limitation. These approaches enable adaptive adjustment of LoRA’s rank, and improve the model’s performance from both local and global perspectives. Lastly, we design detailed algorithms for different adaptation schemes. Extensive experiments demonstrate that our proposed solutions not only effectively improve the performance of the autonomous driving perception model in rare scenarios, but also maintain lower inference latency compared to baseline methods.

In this paper, we present an application-specific Multi-Access Edge Computing (MEC) network architecture by leveraging the Control and User Plane Separation (CUPS) in mobile core networks to offload data processing from central servers to edge servers to reduce the transmitted traffic volume and also the response latency of connected vehicle mobility service. We first apply deep learning to classify packets of different applications to different Radio Access Networks (RAN) slices for application-specific spectrum scheduling. Then, we slice Evolved Packet Core (EPC) and deploy EPC data plane slices on-demand for each application and route packets from RAN slices to edge servers. By applying network slicing, multiple RAN, EPC and MEC slices that support different categories of services with different quality of service (QoS) requirements can be deployed in the same physical infrastructure. We prototype the proposed application-specific CUPS architecture using modified open source software OpenAirInterface on our deeply programmable platform. The preliminary experimental results show the feasibility and efficiency of proposed application-specific CUPS architecture, which can achieve a significant decrease in transmission data volume and latency.