---
license: mit
pipeline_tag: robotics
tags:
- robotics
- world-models
- safety
---

# Modeling Latent Disturbances for Robust Decision-Making in World Models: checkpoints

Pretrained world models and latent safety filters for **LUCID**.
📄 [Paper](https://arxiv.org/abs/2610.07599) · 🌐 [Project page](https://junwon.me/LatentDisturbance/) · 💻 [Code](https://github.com/CMU-IntentLab/LatentDisturbance)

```
checkpoints/
  dubins/          wm_naughty.pt  wm_positional.pt  filter_naughty/  filter_positional/
  block_pouring/   wm.pt  filter/  filter_nominal/  diffusion_policy/
  egg/             wm.pt  norm_stats.json  filter/
```

Download into the root of the code repository, e.g. for the Dubins car:

```bash
hf download junwon-seo/LatentDisturbance --include "checkpoints/dubins/*" --local-dir .
```

The egg world model (`checkpoints/egg/wm.pt`) does not include the frozen DINOv3 ViT-S/16+ encoder; download
`dinov3_vits16plus_pretrain_lvd1689m-4057cbaa.pth` from [DINOv3](https://github.com/facebookresearch/dinov3) and
place it in `checkpoints/egg/`.

Data: [junwon-seo/LatentDisturbance-data](https://huggingface.co/datasets/junwon-seo/LatentDisturbance-data).

## Citation

```bibtex
@article{seo2026latentdisturbance,
  title={Modeling Latent Disturbances for Robust Decision-Making in World Models},
  author={Seo, Junwon and Bajcsy, Andrea},
  journal={arXiv preprint arXiv:2610.07599},
  year={2026}
}
```