Learning to Flock: Evolutionary Emergence of Communication under Fear and Constraint

We propose that collective communication in animals can arise from minimal behavioral pressures. For this, we present a neural-evolutionary framework where minimally intelligent agents evolve under fear-driven cohesion. Each agent adheres to biomechanical and perceptual limits, including maximum turning rate, finite field-of-view, communication noise, and volume exclusion. Despite this simplicity, evolutionary training across 3,000 simulations yields diverse formations, such as flocks, lanes, bands, and segmented swarms. Using Vietoris–Rips filtrations and persistent homology, we construct topological fingerprints of agent interaction and compare the simulations with over 1,000 empirical animal networks across taxa. The results reveal a strong cross-species correspondence: fear-based cohesion alone reproduces both hierarchical and egalitarian, small-world, and near-uniform social topologies. Principal-component analysis identifies communication noise and motility as the primary drivers, followed by reasoning complexity and sensory scope. We conclude that minimal fear and biomechanical constraints are sufficient to generate the cooperative diversity observed across animal societies.