Generative driving world models rely on compact latent state representations that must be efficiently transmitted and synchronized across distributed compute and connected vehicles. We study network-efficient streaming of a discrete world model state, where a stride-16 VQ-U-Net tokenizer (codebook size 8,192) maps each 288x512 frame to an 18x32 grid of token IDs (576 tokens/frame), equivalent to 936 bytes/frame under fixed-length coding. We consider a keyframe–delta protocol under strict per-message payload budgets and packet loss, and propose a fully online, label-free algorithm that prioritizes delta updates via cosine distance in codebook embedding space and triggers keyframes adaptively using a Hamming-drift threshold. The adaptive algorithm consistently improves the rate distortion frontier over periodic keyframes at matched bitrates: at 0.024 Mb/s (200-byte budget) dynamic-only embedding distortion drops from 0.0712 to 0.0661 (7.2%), and at 0.036 Mb/s (400-byte budget) from 0.0427 to 0.0407 (4.8%). Under 10% delta packet loss at 200 bytes, dynamic-only distortion is 0.0757 versus 0.0789 for a matched periodic baseline. To connect state fidelity to world model usefulness, we train a lightweight next-token predictor and evaluate perplexity conditioned on streamed receiver states: at 0.024 Mb/s, dynamic-position perplexity improves from 206.0 to 193.1 (6.3%), and at 0.036 Mb/s from 158.9 to 155.6 (2.1%). These results support discrete token-state streaming as a practical systems layer for bandwidth-aware synchronization and improved downstream token-dynamics utility under vehicular networking constraints.
Vehicular networks face critical connectivity challenges due to the high mobility of their mobile nodes. Multihomed configurations, supported by multiple radio access networks, such as IEEE 802.11p and 4G/5G, are used to complement the different communication ranges, eliminating situations with no connectivity between the vehicles and the infrastructure. On top of this path diversity, higher layer protocols, such as MultiPath Transmission Control Protocol (MPTCP), can be used not only to increase the overall channel capacity, but also to eliminate the effects of horizontal and/or vertical handovers, with the latest one being more critical than the first one.This work proposes and evaluates a solution to dynamically manage MPTCP subflows in multihomed and multiple technology vehicular networks. According to the wireless link conditions, MPTCP subflows are created and removed over the different logical connections between the vehicle and the IEEE 802.11p Roadside Units or the 5G gNodeB. Performance results, obtained through real-world experiments in a city-scale vehicular network, using vehicular network stacks and communication hardware, show that MPTCP can successfully transmit data simultaneously over multiple paths, even when the vehicle is constantly changing its point-of-attachment with the infrastructure. This way, the solution ensures continuous, stable and efficient communication, as well as it improves network performance by combining the different radio access networks.
The development of autonomous vehicles and vehicular communications (V2X) promises to significantly enhance traffic safety and efficiency. However, challenges remain in ensuring safe interactions between autonomous vehicles and vulnerable road users (VRUs). We promote the idea to use a social robot as interface between social interaction and V2X. We propose to use such a robot for crash prediction: the social robot, equipped with an RGB-D camera and V2X-communication capabilities, gathers data on pedestrians’ trajectories and vehicles’ movement within a shared environment. The data can then be used to predict possible crashes. In this ongoing work, we present a framework that integrates the basic functionality to implement such an approach. To test the approach a data set consisting of videos of crossing pedestrians and V2X data of an eBike was collected. The system effectively converts the trajectories of pedestrians and vehicles into a shared coordinate frame, enabling precise detection of potential collisions. The preliminary findings show potential for a novel method for crash prediction.
Future vehicles and infrastructure will rely on data from external entities such as other vehicles via V2X communication for safety-critical applications. Malicious manipulation of this data can lead to safety incidents. Earlier works proposed a trust assessment framework (TAF) to allow a vehicle or infrastructure node to assess whether it can trust the data it received. Using subjective logic, a TAF can calculate trust opinions for the trustworthiness of the data based on different types of evidence obtained from diverse trust sources. One particular challenge in trust assessment is the appropriate quantification of this evidence. In this paper, we introduce different quantification methods that transform evidence into appropriate subjective logic opinions. We suggest quantification methods for different types of evidence: security reports, misbehavior detection reports, intrusion detection system alerts, GNSS spoofing scores, and system integrity reports. Our evaluations in a smart traffic light system scenario show that the TAF detects attacks with an accuracy greater than 96% and intersection throughput increased by 42% while maintaining safety and security, when using our proposed quantification methods.
While huge progress has been made in the field of automated driving in recent years, it becomes more and more clear that certain Operational Design Domains are still a steep challenge for automated vehicles. Hence, teleoperated driving systems are emerging as a solution for situations in which an automated vehicle cannot continue. Most of the systems are based on cameras only and use cellular communication for data transmission. While this is helpful in many scenarios, cameras have limitations, e.g., with fog or at night. In addition, relying on cellular networks can create a point of failure, even with multiple operators. This paper presents an advanced teleoperated driving system that aims to address these issues. The teleoperation system uses LIDAR combined with cameras and allows communication over 5G networks as well as over ITS-G5. The camera view gives a good impression of the scenario, whereas the LIDAR increases the perception of the operator with freespace and object information. Cellular networks can be used when available, and ITS-G5 can be specifically deployed in areas where no cellular signal can be received, for example, underground parkings or stretches of rural highways. The evaluation shows that both communication channels are reliable and fast enough for teleoperation and that different sensor information can be transmitted.
