Latent Diffusion‐Enabled Extreme Road Condition Perception Sharing in Vehicle‐to‐Everything

In vehicle‐to‐everything (V2X) networks, reliable road condition perception sharing is critical for collaborative intelligence under bandwidth constraints and extreme road conditions. However, the captured visual data suffer from semantic structure degradation under extreme environments. Furthermore, these degraded representations are further distorted under complex non‐Gaussian wireless interference, severely undermining semantic perception accuracy. To address these challenges, we propose LD‐S 2 Com, a Latent Diffusion‐driven Semantic Communication framework for Segmentation‐aware perception sharing in V2X. Specifically, we leverage a generative diffusion prior to bridge the gap between corrupted observations and the pre‐learned V2X‐specific semantic manifold. In addition, we introduce a Variational Information Bottleneck (VIB) loss to regularize the learned representation by minimizing task‐irrelevant information, enabling it to prevent the diffusion process from magnifying channel‐induced noise. Experimental results demonstrate that the proposed framework outperforms existing baselines in semantic accuracy and robustness under extreme road conditions with non‐Gaussian interference. © 2026 Institute of Electrical Engineers of Japan and Wiley Periodicals LLC.