Attention-based selective interaction model for self-organized collective motion

This paper proposes an attention mechanism-based selective interaction model to study the self-organized orderly motion of biological groups and their dynamic response behavior to environmental stimuli. By introducing goal-driven (focusing on the most ordered neighbor) and stimulus-driven (focusing on the most anomalous neighbor) attention mechanisms, the model captures velocity differences through visual variations and calculates motion saliency, thereby simulating the adaptive behaviors of biological groups more effectively. Numerical simulation experiments demonstrate that dynamically integrating these two attention mechanisms can significantly improve the polarization metric, with a maximum polarization increase of 72.9 % with a small neighborhood (k = 3) and moderate group size (N = 30). The study also finds that increasing group size reduces the polarization, while increasing the number of topological neighborhoods significantly enhances the polarization; visual angles exhibit a critical threshold (180°) and an optimal range (270°). Furthermore, even in extreme cases with only a single informed individual, the model can effectively guide the response to environmental changes, with the highest average response exceeding 0.88. This research offers a new perspective for modeling the role of selective attention in collective behaviors. The attention mechanisms inform the design of coordination algorithms for artificial systems like robotic swarms.