Three local rules. A whole flock.
Craig Reynolds gave each simulated bird three tiny instincts in 1986 — no leader, no plan, no shared map — and let them loose. From those local urges, a flock emerged. The Explorer lets you tune the three rules in real time and watch the whole choreography ripple out.
In 1986 Craig Reynolds wondered if a flock needs a choreographer. He gave each simulated bird — a «boid» — three local instincts and let them loose. Out of pure neighbour-watching, a flock appears.
Each boid looks at the handful of neighbours inside its perception radius. Separation: steer away from anyone too close. Alignment: rotate toward the mean heading of the group. Cohesion: drift toward the local centre of mass.
Reynolds' rules became the blueprint for emergent behaviour: starlings, fish schools, drone swarms, evacuation simulations, even particle-swarm optimization. Local perception, global pattern — the recipe behind a thousand systems.
Imagine a few hundred triangles, each with a position and a velocity, each able to see only the neighbours inside a small disc around it. No central controller, no shared map, no global goal. Every boid runs the same three-line update rule, every frame, independently — and the flock falls out of the sum.
Look at neighbours that are too close — the ones inside a tighter avoidance radius. Sum a steering vector that points away from each, weighted by the inverse of the distance so closer neighbours push harder. Without separation, the flock collapses to a single point.
Closer neighbours push the focus harder away.
Average the velocities of every neighbour inside the perception radius and steer to match that mean heading. Without alignment, boids cluster but face every which way — there is no «direction of the flock».
Velocity rotates toward the mean heading.
Average the positions of every neighbour inside the perception radius and steer gently toward that local centre of mass. Without cohesion, the boids drift apart — alignment alone cannot keep the flock together.
Steers toward the neighbourhood centroid.
Each triangle is a boid running the same separation-alignment-cohesion update. Nothing is scripted: the streams, splits, and reunions all come from local interactions. Watch the swirls form for a few seconds — every pattern you see is emergent.
Reynolds presented boids at SIGGRAPH 1987; five and seven years later the bats of Batman Returns (1992) and the wildebeest stampede of The Lion King (1994) used the same engine. Today the descendants steer Intel's Shooting Star drone swarms, simulate crowd evacuations, and power particle-swarm optimization — a search algorithm where candidate solutions «flock» toward the best one.
Tune the weights, the perception radius, the avoidance distance. Watch one rule at a time, then add the others back in and see how the flock changes character.
→ Open the ExplorerHow and where this technique lives in the world today.
Topics in the same vein.
