Collective Motion in Spatially Heterogeneous Environments

Abstract

This dissertation investigates how spatial constraints influence collective motion through agent-based modeling and empirical analysis of animal behavior. We study how obstacles affect flocking, developed an agent-based model of bat flight from experimental data, and explored a model-free classification tool for swarming datasets. In the first study, we incorporated physical violation rules to an existing canonical flocking model to understand obstacle-induced perturbations to flocking behavior. The findings revealed that obstacles introduce non-monotonic phases of flocking. This non-monotonicity is tied to flock density, which is in turn mediated by noise and a characteristic length scale - the ratio of sensing radius to the obstacle radius. We found that particles need to sense in the order of the obstacle size to maintain global order. However, a high sensing radius can causes interlocked flocks to collide into virtual obstacles. Hence, while noise in the system generally frustrates global order, it is observed to facilitate flocking by breaking interlocked groups. Next, we extended the study to wildlife systems by analyzing the flight of gray bats. In this second study, we collected experimental data of bats flying in the absence and presence of environmental obstacles, and compared their flight under both conditions. Statistical and agent-based modeling of bat flight revealed that bats shift from spatial-memory driven channel-following behaviors to socially repulsive behavior in constrained spaces. Additionally, we found that obstacle avoidance dominates other interactions, suggesting that environmental feedback supersedes social cohesion under the stress of a novel environment. In the final study, we assessed swarming dynamics from 3D data projected into multiple 2D camera and pixel datasets to examine whether causal information can be retained in low-dimensional and noisy representations. Using midge swarming datasets, the analysis showed that EUGENE, the classification tool employed, performs better in constrained conditions such as cross-wind motion. This result implied that constrained dynamical systems are more sensitive to causal learning and similarity analyses. Overall, the studies in this dissertation reveal that spatial constraints, while limiting freedom of motion, accentuate local interactions and social structures. Counterintuitively, noise (or confusion) and reduced coordination may enhance collective motion in constrained environments.