DeepSC-Empowered Multi-UAV Cooperative Exploration: A Tokenized Approach

Abstract

With the rapid proliferation of the Internet of Everything, the paradigm of multi-UAV cooperative exploration—as a key enabler for Aerial Internet of Things (AIoT)—has emerged as a cornerstone for mission-critical applications, ranging from emergency rescue communications to collaborative sensing in inaccessible environments. Existing studies often neglect the non-stationary nature of UAV wireless links, assuming perfect information exchange while disregarding the severe throughput degradation caused by site-specific fading and constrained communication resources. This paper focuses on reliable multi-UAV cooperative exploration under communication-limited conditions. Unlike traditional bit-level transmission that easily leads to coordination collapse under severe channel impairments, we propose a tokenized semantic communication scheme based on DeepSC to transmit compact and task-relevant information. By prioritizing the exchange of task-essential semantic features, the proposed scheme effectively decouples exploration performance from communication volatility. This enables the UAV swarm to maintain consistent situational awareness and seamless coordination, even when operating in bandwidth-limited and noisy regimes. Numerical evaluations across diverse propagation scenarios, including AWGN and Rayleigh fading environments, demonstrate that the proposed framework achieves superior exploration coverage and exceptional resilience in low-SNR regimes. By consistently out-performing conventional separate source-channel coding benchmarks, our token-based approach proves its ability to preserve coordination fidelity under extreme signal impairments, establishing a robust foundation for autonomous UAV deployments in complex low-altitude missions.