Design a level, play it, then challenge the AI to solve it.
Best on desktop with a keyboard (WASD/arrow keys).
Edit mode: click to paint tiles. Play mode: move with WASD or arrows, jump with space.
Two AI Approaches: Choose between a Genetic Algorithm that evolves action sequences, or a Neural Network that learns to map game state to actions. Both run entirely in your browser with no server required.
Training Process: The AI starts with random behavior and gradually improves through evolution. Each generation tests multiple agents, keeps the best performers, and creates variations through mutation. For neural networks, weights are evolved instead of action sequences.
Fitness & Learning: The AI is rewarded for getting closer to the finish flag and exploring new areas of the level. For challenging levels with large obstacles, an automatic pathfinding system creates intermediate checkpoints to guide the AI.
Deterministic Physics: All game physics use fixed timesteps and deterministic calculations, ensuring that the same actions always produce the same results. This makes training reliable and reproducible.