SandboxLabML

Train your own model with machine learning, right in your browser.

Or play Hide and Seek, or write your own rules in the Studio.

How it works

1
Pick a brain
Choose what the agents sense, from 3 rays to 60 inputs.
2
Write the rules
Rewards and how each generation breeds, as blocks or code.
3
Watch it evolve
NEAT breeds the best brains and grows their networks over time.
4
Measure it
Benchmark the champion on tracks it never saw.

Braking and turning share the same tire grip, so cars must learn to brake before corners.

Past champions drive as ghosts beside the live cars, so you see the laps get faster.

Hiders lock boxes into forts. Seekers push ramps to jump the walls.

Write the training rules as blocks or code. Both views edit the same script.

Pick what the agents sense and watch every neuron fire live.

Short lessons that build a training script step by step.

Scores a champion on tracks it never trained on, so runs compare fairly.

Race your champions on a track you draw, or drop them in a room you build.