Navigation World Models, CVPR 2025 (Oral)
_{Official PyTorch Implementation}

Paper | Project Page | Notebook Demo | Models

This repo contains the official PyTorch implementation of Navigation World Models- the Conditional Diffusion Transformer (CDiT) model training code. See the project page for additional results.

Navigation World Models
Amir Bar, Gaoyue "Kathy" Zhou, Danny Tran, Trevor Darrell, Yann LeCun
AI at Meta, UC Berkeley, New York University

Setup

First, download and set up the repo:

git clone https://github.com/facebookresearch/nwm
cd nwm

Data

To download and preprocess data, please follow the steps from NoMaD, specifically:

• Download the datasets
• Change the preprocessing resolution from (160, 120) to (320, 240) for higher resolution
• run process_bags.py and process_recon.py to save each processed dataset to path/to/nwm_repo/data/<dataset_name>.

For SACSon/HuRoN, we used a private version which contains higher resolution images. Please contact the dataset authors for access (we're unable to distribute).

Finally, you should have the following structure:

nwm/data
├── <dataset_name>
│   ├── <name_of_traj1>
│   │   ├── 0.jpg
│   │   ├── 1.jpg
│   │   ├── ...
│   │   ├── T_1.jpg
│   │   └── traj_data.pkl
│   ├── <name_of_traj2>
│   │   ├── 0.jpg
│   │   ├── 1.jpg
│   │   ├── ...
│   │   ├── T_2.jpg
│   │   └── traj_data.pkl
│   ...
└── └── <name_of_trajN>
    	├── 0.jpg
    	├── 1.jpg
    	├── ...
        ├── T_N.jpg
        └── traj_data.pkl

Requirements:

mamba create -n nwm python=3.10
mamba activate nwm
pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu126
mamba install ffmpeg
pip3 install decord einops evo transformers diffusers tqdm timm notebook dreamsim torcheval lpips ipywidgets

Training

Using torchrun:

export NUM_NODES=8
export HOST_NODE_ADDR=<HOST_ADDR>
export CURR_NODE_RANK=<NODE_RANK>

torchrun \
  --nnodes=${NUM_NODES} \
  --nproc-per-node=8 \
  --node-rank=${CURR_NODE_RANK} \
  --rdzv-backend=c10d \
  --rdzv-endpoint=${HOST_NODE_ADDR}:29500 \
  train.py --config config/nwm_cdit_xl.yaml --ckpt-every 2000 --eval-every 10000 --bfloat16 1 --epochs 300 --torch-compile 0

Or using submitit and slurm (8 machines of 8 gpus):

python submitit_train_cw.py --nodes 8 --partition <partition_name> --qos <qos> --config config/nwm_cdit_xl.yaml --ckpt-every 2000 --eval-every 10000 --bfloat16 1 --epochs 300  --torch-compile 0

Or locally on one GPU for debug:

python train.py --config config/nwm_cdit_xl.yaml --ckpt-every 2000 --eval-every 10000 --bfloat16 1 --epochs 300  --torch-compile 0

Note: torch compile can lead to ~40% faster training speed. However, it might lead to instabilities and inconsistent behvaior across different pytorch versions. Use carefuly.

Pretrained Models

To use a pretrained CDiT/XL model:

• Download a pretrained model from Hugging Face
• Place the checkpoint in ./logs/nwm_cdit_xl/checkpoints

Evaluation

directory to save evaluation results: export RESULTS_FOLDER=/path/to/res_folder/

Evaluate on single time step prediction

1. Prepare ground truth frames for evaluation (one-time)

python isolated_nwm_infer.py \
    --exp config/nwm_cdit_xl.yaml \
    --datasets recon,scand,sacson,tartan_drive \
    --batch_size 96 \
    --num_workers 12 \
    --eval_type time \
    --output_dir ${RESULTS_FOLDER} \
    --gt 1

2. Predict future state given action

python isolated_nwm_infer.py \
    --exp config/nwm_cdit_xl.yaml \
    --ckp 0100000 \
    --datasets <dataset_name> \
    --batch_size 64 \
    --num_workers 12 \
    --eval_type time \
    --output_dir ${RESULTS_FOLDER}

3. Report metrics compared to GT (LPIPS, DreamSim, FID)

python isolated_nwm_eval.py \
    --datasets <dataset_name> \
    --gt_dir ${RESULTS_FOLDER}/gt \
    --exp_dir ${RESULTS_FOLDER}/nwm_cdit_xl \
    --eval_types time

Results are saved in ${RESULTS_FOLDER}/nwm_cdit_xl/<dataset_name>

Evaluate on following ground truth trajectories

1. Prepare ground truth frames for evaluation (one-time)

python isolated_nwm_infer.py \
    --exp config/nwm_cdit_xl.yaml \
    --datasets recon,scand,sacson,tartan_drive \
    --batch_size 96 \
    --num_workers 12 \
    --eval_type rollout \
    --output_dir ${RESULTS_FOLDER} \
    --gt 1 \
    --rollout_fps_values 1,4

2. Simulate a GT trajectory using NWM

python isolated_nwm_infer.py \
    --exp config/nwm_cdit_xl.yaml \
    --ckp 0100000 \
    --datasets <dataset_name> \
    --batch_size 64 \
    --num_workers 12 \
    --eval_type rollout \
    --output_dir ${RESULTS_FOLDER} \
    --rollout_fps_values 1,4

3. Report metrics compared to GT trajectories (LPIPS, DreamSim, FID)

    python isolated_nwm_eval.py \
        --datasets recon \
        --gt_dir ${RESULTS_FOLDER}/gt \
        --exp_dir ${RESULTS_FOLDER}/nwm_cdit_xl \
        --eval_types rollout

Results are saved in ${RESULTS_FOLDER}/nwm_cdit_xl/<dataset_name>

Trajectory Evaluation - Planning

Using 1-step Cross Entropy Method planning on 8 gpus (sampling 120 trajectories):

torchrun --nproc-per-node=8 planning_eval.py \
    --exp config/nwm_cdit_xl.yaml   \
    --datasets recon   \
    --rollout_stride 1   \
    --batch_size 1   \
    --num_samples 120   \
    --topk 5   \
    --num_workers 12   \
    --output_dir ${RESULTS_FOLDER}   \
    --save_preds   \
    --ckp 0100000   \
    --opt_steps 1   \
    --num_repeat_eval 3

Results are saved in ${RESULTS_FOLDER}/nwm_cdit_xl/<dataset_name>

BibTeX

@article{bar2024navigation,
  title={Navigation world models},
  author={Bar, Amir and Zhou, Gaoyue and Tran, Danny and Darrell, Trevor and LeCun, Yann},
  journal={arXiv preprint arXiv:2412.03572},
  year={2024}
}

Acknowledgments

We thank Noriaki Hirose for his help with the HuRoN dataset and for sharing his insights, and to Manan Tomar, David Fan, Sonia Joseph, Angjoo Kanazawa, Ethan Weber, Nicolas Ballas, and the anonymous reviewers for their helpful discussions and feedback.

License

The code and model weights are licensed under Creative Commons Attribution-NonCommercial 4.0 International. See LICENSE.txt for details.