Goal-Oriented Semantic Twinning-Enabled Collaborative Target Tracking in Communication-Constrained Satellite-UAV Networks

In satellite and uncrewed aerial vehicle (UAV) networks, dynamic network topology, unstable channels, and distributed computing resources severely degrade collaborative tracking performance under communication constraints. This paper presents a goal-oriented semantic twinning (GOST) system that enables globally accurate perception and efficient collaborative decision-making through on-demand modeling and semantic transmission. To counter data staleness from high latency and packet loss, we design a multidimensional data inference mechanism exploiting temporal, kinematic, spatial, and causal features. Leveraging the global perspective of GOST, we develop a satellite UAV collaborative decision-making framework based on multi agent deep deterministic policy gradient (MADDPG) algorithm, with incremental learning via elastic weight consolidation (EWC) and sample-weighted replay for dynamic adaptation. Simulation results demonstrate that GOST reduces the age of information (AoI) by 68% and positioning error by 75% compared to conventional digital twin (DT) approaches under communication constrained channels, achieving a 50% reduction in the target loss, and improves sample efficiency by 66% when the target motion pattern changes, thereby significantly enhancing robustness.