OpenSpace Lab
Read paper

OpenNWM

A Generalist Navigation World Model
with Latent Action Pretraining

Read the paper Code Model Try the demo Dataset · NavAnywhere
NavAnywhere
EXPLORE OPENNWM
  • Dujun Nie1,2*
  • Ruiqi Song1,6*
  • Zixuan Wang1,2
  • Hanqing Wang3
  • Yifan Han1,2,4
  • Baiyong Ding1
  • Wenzhao Zheng5
  • Hangbin Wu6
  • Long Chen1†
  • 1CASIA
  • 2UCAS
  • 3HKUST (Guangzhou)
  • 4Shanghai Jiao Tong University
  • 5UC Berkeley
  • 6Tongji University

* Equal contribution† Corresponding author

01 / OVERVIEW

Overview of OpenNWM

OpenNWM method: a latent action encoder and decoder infer actions from observations; a conditional diffusion transformer predicts future frames with frozen VAE encoding and decoding.
02 / MAIN RESULTS

Future prediction & navigation

04 / VISUALIZATION

Visualization

Predicted views along given paths, followed by offline plans toward visual goals.

Future prediction

05 / Demo

Your path.
An imagined future.

Choose a scene and a path. Let OpenNWM imagine the view ahead.

Checking live service…
01 — DRAW YOUR PATH
FORWARD ↑Drag from the starting point
02 — IMAGINE THE VIEWSTARTING VIEW
Selected scene's initial observation
A new journey starts here

06 / NavAnywhere Dataset

Different worlds.
A common language.

Streets, homes, gardens, and paths less traveled. Navigation experience across people, robots, and drones.

Explore the dataset
17.5Mvisual observations
15pretraining sources
70.8Knavigation sequences
KEEP EXPLORING

The next step
starts here.

A Generalist Navigation World Model with Latent Action Pretraining

Visit OpenSpace LabRead the paperProject code
REFERENCE THIS WORK 
@misc{nie2026opennwmgeneralistnavigationworld,
      title={OpenNWM: A Generalist Navigation World Model with Latent Action Pretraining},
      author={Dujun Nie and Ruiqi Song and Zixuan Wang and Hanqing Wang and Yifan Han and Baiyong Ding and Wenzhao Zheng and Hangbin Wu and Long Chen},
      year={2026},
      eprint={2611.01844},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2611.01844},
}

arXiv:2611.01844 · 2026.