Motion topology inference and trajectory prediction method for unmanned swarm system

Motion topology inference and trajectory prediction of unmanned swarm systems are the core challenges of autonomous coordination and game confrontation. The existing data-driven methods often neglect physical constraints, leading to prediction error accumulation, limited interpretability, and inadequate motion topology reconstruction. To solve this problem, this paper proposes Swarm Relational Inference (SRI), an unsupervised end-to-end prediction model that integrates swarm dynamics with dynamic graph neural networks to achieve interpretable high-precision trajectory prediction through motion topology inference. Firstly, a physically interpretable motion topology graph is constructed by encoding swarm dynamics into explicit edge types and node state. Secondly, node states and dynamic relationships are jointly represented through multi-dimensional Long Short-Term Memory (LSTM) networks and multi-head attention mechanism, enabling real-time motion topology inference. Finally, a variational inference framework jointly optimizes topology inference and trajectory prediction using only historical trajectory inputs, establishing an interpretable prediction mechanism. Experiments show that SRI is significantly better than the existing methods in terms of topology inference and trajectory prediction accuracy, effectively reveals the dynamic evolution mechanism of formation structure, and can learn the implicit topology and understand the swarm intent in complex scenes with unclear motion topology, which provides an interpretable and high-precision theoretical tool for swarm cooperative control and behavior understanding.