PROCEDURAL PLAYGROUND
Generate a challenge
Seed: —
—
YOUR CONTROLLER
Neural network weights
Each motor uses tanh(Σ input × weight + bias). Change any weight, then test your model. Inputs are normalized to roughly −1 to +1.
The example controller is a starting point, not an optimized solution. For Hopper, try to coordinate hip rotation, foot contact, and leg extension. Submitted weights are frozen during evaluation.
UNLIMITED PRACTICE
Test your model
Edit weights and start a practice run.
FINAL EVALUATION
Three rounds, one controller
1 · RandomLocked
Waiting for submission
2 · BasicLocked
Beat round 1 to unlock
3 · StrongerLocked
Beat round 2 to unlock