Deriving Low-Level Steering Behaviors from Trajectory Data

Emergent behavior, such as flocks and swarms appears in numerous multi-agent systems in nature. Such behaviors emerge not through centralized high-level control, but through low-level local interactions between each agent and its immediate environment. The understanding of individual local interactions between agents within a group is therefore essential for the understanding of emergent group behaviors. The focus of recent work has been primarily on developing tools for the detection and mining of group behaviors (e. g., spatiotemporal clusters), without offering the ability to link such behaviors to individual agent behavior. Focusing on steering behaviors, this work aims to address this gap by developing a methodology for estimating agent steering behaviors that would explain the emergent group behavior observed in trajectory data. In particular, we present a particle swarm optimization-based tracking scheme for deriving agent steering behaviors based on Reynolds´ boids model. The paper formally outlines the low-level agent behavior derivation problem and discusses our proposed methodology. In addition, results from implementing our approach on real-world data are presented.