ImagiNav: Scalable Embodied Navigation via Generative Visual Prediction and Inverse Dynamics
Paper • 2603.13833 • Published
This dataset supports the paper ImagiNav: Scalable Embodied Navigation via Generative Visual Prediction and Inverse Dynamics.
The dataset contains processed egocentric navigation clips from real-world and simulation sources, with motion annotations, used for training and evaluating the ImagiNav video generation and navigation pipeline.