OpenNWM: A Generalist Navigation World Model with Latent Action Pretraining
OpenNWM: A Generalist Navigation World Model with Latent Action Pretraining
This research presents OpenNWM, a generalist navigation world model that leverages latent action pretraining to enhance generalization across real-world environments. The authors introduce NavAnywhere, a large-scale dataset comprising 17.5 million visual observations from 15 heterogeneous sources, capturing navigation by ground robots, humans, and drones across five broad scene types. By learning universal latent actions from action-free videos and aligning them with physical motions through post-training, OpenNWM achieves improved visual prediction and navigation performance on both in-domain and out-of-domain benchmarks.
The paper introduces OpenNWM, a navigation world model that uses latent action pretraining on NavAnywhere—a dataset of 17.5 million action-free video observations from 15 sources—to improve generalization across diverse environments.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
visual observations from sources
15
View original evidence
vations from 15 heterogeneous sources. It captures navigation by ground robots, humans, and drones, with diverse viewpoints and motion patterns across five broad scene types: residential areas, public and commercial spaces, urban settings, parks, and natural landscapes. Experiments demonstrate improved visual prediction on in-domain and out-of-domain benchmarks, together with improved navigation performance. These re
Open source S4sources
15
View original evidence
vations from 15 heterogeneous sources. It captures navigation by ground robots, humans, and drones, with diverse viewpoints and motion patterns across five broad scene types: residential areas, public and commercial spaces, urban settings, parks, and natural landscapes. Experiments demonstrate improved visual prediction on in-domain and out-of-domain benchmarks, together with improved navigation performance. These re
Open source S4scene types
5
View original evidence
vations from 15 heterogeneous sources. It captures navigation by ground robots, humans, and drones, with diverse viewpoints and motion patterns across five broad scene types: residential areas, public and commercial spaces, urban settings, parks, and natural landscapes. Experiments demonstrate improved visual prediction on in-domain and out-of-domain benchmarks, together with improved navigation performance. These re
Open source S4
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
ntroduce OpenNWM, a generalist navigation world model with latent action pretraining. It learns universal latent actions from large-scale action-free videos through visual prediction pretraining, and further aligns them with physical motions via action-conditioned post-training. To support video pretraining, we curate NavAnywhere, a large-scale action-free navigation video dataset comprising 17.5 million visual obser
Open source S3
vations from 15 heterogeneous sources. It captures navigation by ground robots, humans, and drones, with diverse viewpoints and motion patterns across five broad scene types: residential areas, public and commercial spaces, urban settings, parks, and natural landscapes. Experiments demonstrate improved visual prediction on in-domain and out-of-domain benchmarks, together with improved navigation performance. These re
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org