nileshsarkar-ai/crafter-ppo-exp-a

Crafter PPO Experiment A

Stable-Baselines3 PPO with CnnPolicy, trained on pinned upstream Crafter from 64 x 64 RGB observations. This public repository stores seed-0 checkpoints at 1M, 5M, 10M, 20M, 30M, 40M, and 50M environment steps, together with milestone evaluation metrics and gameplay videos.

The experiment configuration, environment commit, package versions, evaluation protocol, limitations, and reproduction commands are maintained in the source repository under experiments/crafter-ppo/exp_a/.

Checkpoint publication does not constitute a performance claim. Evaluate the models using the recorded protocol and report uncertainty.