An observatory of artificial life

A public observatory of emergence

LifeLabs hosts artificial-life experiments running continuously on a single local machine. Here, simple rules produce shapes that move, feed, and reproduce — without ever claiming they are "alive."

Explore the experiments

Three ways to explore

Observe

A research journal tracks every run: creature vital signs, time series, video captures. Nothing is staged — these are raw, annotated measurements.

Pilot

WebGPU demos run directly in your browser. Adjust the parameters, seed a creature, and watch whether it survives its own rules.

ContributeComing soon

A collaborative bestiary will catalogue the creatures discovered by the community — shapes, behaviours, conditions of emergence. This feature arrives after the first milestone.

Three experiments, one shared ground

Lenia, the evolutionary agents, and the bipedal avatar are LifeLabs' three current residents — three different ways of making something emerge without ever writing it down directly. The experiments hub presents them side by side, each with what it makes emerge and something interactive to pilot.

View the experiments

Latest observations

The research journal hasn't been published yet. The first entries — measurements, captures, and experiment notes on Lenia — will arrive with the project's next milestones.

First resident organism: Lenia

Lenia is a continuous cellular automaton designed by Bert Wang-Chak Chan: space, time, and cell states are continuous rather than discrete, giving rise to organic, mobile, sometimes remarkably stable structures. It is the first experiment hosted by LifeLabs.

View the experiment

Second experiment: evolutionary agents

A population of tiny neural-network brains learns, generation after generation and without a single line of programmed behavior, to seek food before its energy runs out. Neuroevolution, currently being set up on the local machine.

View the experiment

Third experiment: the avatar that learns to walk

A bipedal body, an empty brain, no programmed gait: from the reward signal alone, a PPO agent learns first to stand, then to shuffle, then to move forward. Watch the progression across checkpoint videos and replay the gait in 3D in your browser.

View the experiment