Reproduce “OpenNWM: A Generalist Navigation World Model with Latent Action Pretraining” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Dujun Nie, Ruiqi Song, Zixuan Wang, Hanqing Wang, Yifan Han, Baiyong Ding, Wenzhao Zheng, Hangbin Wu and 1 more
OpenNWM trains a navigation world model by learning latent motion codes from visual transitions in action-free video, grounding them with physical-motion annotations where available, and adapting the predictor to real waypoint actions. It reports a four-dataset in-domain 4-second average LPIPS of 0.368, versus NWM’s 0.405, and lower Office-Go2 ATE/RPE than RAE-NWM; results are limited to evaluated benchmarks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.