VAE-Driven Semantic Communication for Autonomous Vehicles.

The evolving landscape of smart transportation and 6G wireless technology demands highly efficient, reliable, and low-latency communication for connected autonomous vehicles (CAVs), especially for safety-critical functions like traffic sign recognition. Traditional communication systems are inefficient in resource-constrained satellite channels because they transmit all raw data, regardless of its relevance to the task. Semantic communication offers a solution by extracting and transmitting only the essential, task-relevant information. This research introduces a Variational Autoencoder (VAE)-based multi-task semantic communication framework specifically designed for satellite-assisted autonomous driving. Unlike deterministic autoencoder methods, this model uses probabilistic latent representations, enhancing robustness and encoding efficiency. The framework is trained end-to-end to jointly optimize both traffic sign reconstruction and classification tasks. Results demonstrate a substantial bandwidth reduction, ranging from 87.23% to 98.17%, while consistently maintaining stable performance across varying signal-to-noise ratio conditions. This approach is crucial for optimizing communication in bandwidth-scarce and high-loss satellite environments.