Dreamer-CDP: Improving Reconstruction-free World Models Via Continuous Deterministic Representation Prediction

We first thank Danijar Hafner for the release of DreamerV3.

Dreamer-CDP

Dreamer-CDP learns a world model without reconstruction through continuous deterministic representation prediction. It reaches similar performance as the reconstruction-based Dreamer-V3 on the Crafter environment. Link to the paper:

Instructions

The code has been tested on Linux and requires Python 3.11+.

Docker

You can either use the provided Dockerfile that contains instructions or follow the manual instructions below.

Manual

Install JAX and then the other dependencies:

pip install -U -r requirements.txt

Training script:

python dreamerv3/main.py \
  --logdir ~/logdir/dreamer/{timestamp} \
  --configs crafter \
  --run.train_ratio 32

If you find this code useful, please reference in your paper:

@misc{hauri2026dreamercdp,
      title={Dreamer-CDP: Improving Reconstruction-free World Models Via Continuous Deterministic Representation Prediction}, 
      author={Michael Hauri and Friedemann Zenke},
      journal={arXiv preprint arXiv:2603.07083}
}