HAPPD: Cooperative Pursuit Strategy for Fixed-Wing UAVs in Complex Terrain

This paper addresses the challenge of perceptual interruptions during fixed-wing UAVs’ pursuit of high-speed targets in complex terrain. These interruptions, caused by persistent terrain occlusion and the evader’s speed advantage, often lead to the failure of conventional strategies under partial observability. To overcome this, we propose a novel Hierarchical Adaptive Perception–Prediction–Decision (HAPPD) framework. The perception level dynamically identifies three occlusion modes. Based on the perception output, the prediction level adaptively activates a dual-layer trajectory prediction module: the upper layer predicts the flight path angle, while the lower layer estimates immediate control actions. The decision level integrates a curriculum learning mechanism and adversarial training to effectively leverage predictive information, enhancing the adaptability and robustness of the pursuit policy. Through synergistic interaction among these components, the framework maintains reliable state estimation and effective pursuit capability even during persistent target occlusion. Simulation results demonstrate that HAPPD achieves an 80.2% success rate in a canyon terrain, significantly outperforming baseline methods. Extensive quantitative analysis further validates the effectiveness of each proposed module.