A Transformer Self-Attention Guided Graph Convolutional Network for Drone Swarm Trajectory Prediction
Qi Li, Jianxaing Xi, Xiaogang Yang, Ruitao Lu, Xueli Xie · 2024
This paper presents a novel trajectory prediction algorithm for infrared drone swarms, addressing the challenges posed by flexible motion patterns and complex intraswarm interactions. The proposed method leverages a Transformer-based self-attention mechanism within a graph attention convolutional network framework. In the encoding phase, a graph convolutional neural network (GCN) is employed to enhance the representation of temporal features. Subsequently, an attention-guided graph neural network is utilized to effectively capture and encode interaction features among drones within the swarm. The Transformer attention mechanism is then integrated into the decoder to refine the prediction process. This encoder-decoder architecture is specifically designed to extract and map the temporal and spatial interaction features of individual drones, providing accurate trajectory predictions that consider the unique dynamics of drone swarms. Experimental results on an infrared drone swarm trajectory prediction dataset, as well as the publicly available pedestrian trajectory prediction datasets ETH/UCY, demonstrate that the proposed algorithm outperforms state-of-the-art methods, achieving superior performance in terms of ADE and FDE metrics.