The official implementation of 'Self-supervised Hierarchical Visual Reasoning with World Model' (OpenReview link), a model-based RL algorithm with self-supervised hierarchical visual reasoning representation.

Instructions

The code has been tested on Ubuntu 20/24 and Python 3.11. Nvidia GPU with at least 24 GB VRAM is recommended for training.

Setup

Install Minedojo, MineClip and the other dependencies:

pip install -U -r requirements.txt

Training

python dreamerv3/main.py \
  --logdir ~/logdir/dreamer/{timestamp} \
  --configs minedojo 

Disclaimer

This repository is based on the open source DreamerV3 code base https://github.com/danijar/dreamerv3.