E.VOLV evolves populations of simple stick creatures to walk, jump, or climb using a genetic algorithm layered on 2D physics — a small-scale homage to Karl Sims' 1994 research Evolving Virtual Creatures. Nobody designs the gait by hand: set a goal and press start, and locomotion emerges purely from random mutation and survival of the fittest. The bold creature is this generation's champion; the faint trails behind it are every other attempt in the population. See About below for more.
What this is. E.VOLV is a genetic algorithm creature evolver: a population of simple two-dimensional stick creatures is randomly generated, then evolved — generation after generation — to get better at a chosen physical task, using nothing but random mutation, sexual reproduction, and survival of the fittest. No neural network, no hand-tuned gait — the walking (or jumping, or climbing) you see was never programmed, only selected for.
Where the idea comes from. This project is a small-scale homage to Karl Sims' 1994 research Evolving Virtual Creatures, which used genetic algorithms and physics simulation to evolve both the bodies and behavior of 3D creatures. E.VOLV simplifies that to 2D stick figures running in a browser tab, but the core idea is the same.
How a generation works. Every creature in the population attempts the current goal at once, each simulated independently for a few seconds. Whichever creature travels farthest (walk/jump) or climbs highest (climb) becomes that generation's "champion." The next generation is bred from the fittest creatures: two parents combine their genes (weighted toward whichever parent performed better), the result is mutated slightly, and a few top performers ("elites") carry over unchanged so progress is never lost.
What evolves. Each creature's genome encodes how its legs move (the rhythm, range, and timing of each leg's swing), how big each leg is, and how many legs it has (2 to 6) — so body shape changes over generations too, not just gait.