AI Platformer Trainer

Este projeto usa algoritmo genetico e rede neural para aprender a jogar um mini platformer, mantendo o restante da estrutura voltado para treino, observacao e experimentacao.

Network View

Inputs, hidden layer and outputs are updated for the current agent.

Status

Mode: Training
Generation: 0
Agent: 0 / 0
Frames: 0
Best Fitness: 0
Best Finish: 0%
Gen Best: 0
Current Fitness: 0
Completion: 0%

History

0 0

O grafico mostra o melhor fitness de cada geracao ja concluida.

Live Brain

Solid Cells: 0
Danger Cells: 0
Player Tile: 0
Grounded: No
Velocity X: 0
Velocity Y: 0
Output Left: 0
Output Right: 0
Output Jump: 0
Action: Idle

Controls

Training Speed

Simulation steps per render

40 steps

Turbo mode ignores the slider and runs headless training much faster.

How It Works

Inputs: uma grade local 9x9 ao redor do jogador, marcada com blocos solidos e vazio perigoso. Outputs: move left, move right, jump.

In manual mode, use Left, Right and Space. Press R to restart.