← back Neural Particle Automata

Origin

Zhu et al., Neural Particle Automata, arXiv 2601.16096, January 2026. Lagrangian generalization of Neural Cellular Automata. Each particle has continuous position and internal state, both updated by a shared learned rule. This demo uses hand-tuned interaction rules inspired by the paper's discovered dynamics.

How It Works

Particles perceive neighbors within a radius, computing a local field from relative positions and states. Position and state updates are computed from this perception. Asynchronous Bernoulli sampling means not all particles update every frame.

Applications

Morphogenesis, point-cloud self-organization, particle-based texture synthesis. Demonstrates that NCA principles extend to Lagrangian particle systems.

choose a preset target behavior·adjust perception radius and update rate·random to scatter particles