Abstract
Swarming systems, such as drone fleets and robotic teams, exhibit complex dynamics driven by both individual behaviors and emergent group-level interactions. Unlike traditional multi-agent domains such as pedestrian crowds or traffic systems, swarms typically consist of a few large groups with inherent and persistent memberships, making group identification essential for understanding fine-grained behavior. We introduce the novel task of group prediction in overlapping multi-agent swarms, where latent group structures must be inferred directly from agent trajectories without ground-truth supervision. To address this challenge, we propose SIGMAS (Second-order Interaction-based Grouping for Multi-Agent Swarms), a self-supervised framework that goes beyond direct pairwise interactions and model second-order interaction across agents. By capturing how similarly agents interact with others, SIGMAS enables robust group inference and adaptively balances individual and collective dynamics through a learnable gating mechanism for joint reasoning. Experiments across diverse synthetic swarm scenarios demonstrate that SIGMAS accurately recovers latent group structures and remains robust under simultaneously overlapping swarm dynamics, establishing both a new benchmark task and a principled modeling framework for swarm understanding.
Abstract (translated)
集群系统，例如无人机群和机器人团队，展现出由个体行为以及群体层面的涌现性交互共同驱动的复杂动态。与传统的多智能体领域（如行人人群或交通系统）不同，集群通常包含少数几个具有固有且持久成员身份的大组，因此识别这些小组对于理解精细粒度的行为至关重要。我们引入了重叠多智能体群集中的群体预测这一新任务，在该任务中需要直接从智能体轨迹推断出潜在的群体结构，而无需地面真实监督。 为了应对这个挑战，我们提出了SIGMAS（基于二阶互动的多智能体集群分组），这是一个自监督框架，它超越了直接成对交互，同时建模智能体之间的二阶互作。通过捕捉代理如何与他人相互作用的方式，SIGMAS能够实现稳健的群体推断，并通过一个可学习的门控机制进行联合推理来适应性地平衡个体和集体动态。 跨多种合成集群场景的实验表明，SIGMAS可以准确恢复潜在的群组结构，并在同时重叠的群集动力学下保持鲁棒性。这不仅建立了新的基准任务，还为群集理解提供了一个原则性的建模框架。
URL
https://arxiv.org/abs/2603.00120
