This scene runs a genuine Particle Swarm Optimization (PSO) search, one of the classic bio-inspired algorithms modeled on how flocks of birds and schools of fish converge on food sources. Each glowing sphere is a candidate solution (x, y) flying above a bumpy multi-peak landscape z = f(x, y). Every particle remembers the best spot it personally found (its "personal best", shown as a small dot) and the whole swarm shares the single best spot anyone has found (the "global best", marked with a pulsing beacon).
At every iteration each particle's velocity is nudged by three forces: its own momentum, an attraction back toward its personal best, and an attraction toward the swarm's global best. Height on the terrain directly encodes fitness, so you can watch the swarm literally climb toward the tallest peak.
v_i(t+1) = w * v_i(t)
+ c1 * r1 * (pbest_i - x_i(t))
+ c2 * r2 * (gbest - x_i(t))
x_i(t+1) = x_i(t) + v_i(t+1)
r1, r2 ~ Uniform(0,1) f(x,y) = landscape height
- Run / Step / Reset — play the search continuously, advance one iteration at a time, or reseed the swarm at random positions.
- Inertia w — how much old velocity survives each step; high w explores broadly, low w lets the swarm settle quickly.
- Cognitive c1 — pull toward each particle's own best-ever spot (individual memory / exploitation of personal history).
- Social c2 — pull toward the swarm's shared best spot (social communication, the "swarm intelligence" term).
- Swarm size — number of birds searching in parallel; more particles cover the landscape faster but cost more computation.
The same velocity/position update rule is used in real engineering to tune neural network weights, antenna designs and logistics routes — anywhere a good-enough answer must be found in a huge, bumpy search space without computing gradients.