← Catalog
PaperMeta / New York University · 2024
Navigation World Models
A controllable video world model for navigation that imagines future observations from planned trajectories and can be used for planning.
Summary
Introduces a conditional diffusion transformer that predicts future egocentric observations given past frames and navigation actions, trained across agents and environments. It supports trajectory planning by simulating candidate paths and can imagine unseen environments from a single image.
Metadata
- Authors
- Bar et al., Meta / NYU
- Year
- 2024
- arXiv
- 2412.03572
- Venue
- CVPR 2025
Relationships
Introduces
Navigation World Model
#navigation#video#planning