Aware of Physics learns local rules behind AI flock and mass movement behavior

Researchers at Seoul National University and King Hei University report a framework to control collective movements, such as colors, clips, mills, herds,, which is informed of physics to learn local rules that follow the interactions among individuals.

This dissertation has appeared in the journal Cell reports physical science.

The approach clarifies when an ordered state should show geometric features (average radius, cluster size, herd size) from random preliminary conditions and gestures. In addition, trained on real pigeon -published GPS tractors, the model expose the procedure of interaction observed in the real herd.

The collective movement is an emerging trend in which many self -driven people (birds, fish, insects, robots, even human crowds) produce large -scale samples without any central decision -making. Each person reacts only to the nearest neighboring countries, yet the group exhibits integrated collective movement. To analyze how such global discipline is born with simple local interactions is a challenge because these systems are noisy and non -liner, and the idea is often difficult.

Learn local rules with AI aware of physics

To tackle these challenges, the team created nervous networks that comply with dynamics laws and are trained on easy patterns and, when available, experimental paths.

The nerve networks evaluate the two basic principles of local interactions: distance -based rules that align, speed -based rules that align the titles, as well as compose their combination. The team also showed that the self -powered agents who follow these rules reiterate the desired target samples with specific geometric features.

Examples include adjusting color radius, cluster size in cluster sizes, and adjusting the rotation mode (either single or double). Adding continuous transition to different collective methods; And the pursuit of motivations in the near obstacles and in limited areas.

The same framework can fit for the short -class real -paced sections, by adding an anacteropic field theory, which enjoys the rules of interaction according to the ranking of the leader, which is observed in nature.

Open new possibilities in collective behavior and robotics

This approach offers practical engineering and scientific benefits by converting collective behavior into something that can be ruled out. In robotics, it provides a blueprint for the programming drone and the ground robot crowd to create and switch samples on demand.

In natural sciences, it helps to identify what local interactions are sufficient to explain the observed herd, which can be examined by speculation about sensory boundaries and alignment strength.

More widely, this procedure can guide the design of active materials that combine itself in target shapes and can help produce realistic artificial datasters to study complex, decentralized systems.