Robotics Model Reveals How Simple Sensorimotor Rules Produce Collective Animal Motion

For decades, collective animal behavior like flocking and schooling has been modeled with abstract interaction forces — attraction, repulsion, alignment — that reproduce patterns but don't explain how individual organisms actually perceive and act. A new paper from Vito Mengers and colleagues at TU Berlin's Robotics and Biology Laboratory tackles this gap head-on. They propose a mechanistic model built from first-person sensorimotor regularities: each agent perceives neighbors only through bearing and apparent-size cues within a limited field of view, maintains uncertain internal state estimates, and selects actions by gradient descent on a desired social distance. There are no prescribed forces. The model is grounded in what a real organism (or robot) can actually sense and compute.

The results are striking. Despite its simplicity, the model generates diverse collective behaviors: polarized motion (flocking), milling (rotation around a center), ring formations, and subgroup fragmentation. A global sensitivity analysis reveals that transitions between these behaviors are governed by sensorimotor parameters that map directly to measurable biological quantities: field-of-view width, sensory noise levels, turning agility, and memory decay rate. This reframes collective motion as an emergent outcome of embodiment and environment, not top-down rules. The implications extend to swarm robotics: instead of hand-crafting interaction laws, engineers could design robots with the right sensorimotor constraints and let collective behavior emerge. The paper is available on arXiv and currently under review.