ATLAS: Aligned Transport of Latent Structure for Reliable World Model Planning
ATLAS: Aligned Transport of Latent Structure for Reliable World Model Planning
This research introduces ATLAS, a method that preserves relational geometry in latent representations for reliable world model planning. By calibrating the global latent distribution through Wasserstein embedding matching, ATLAS improves performance on tasks like PushT, TwoRoom, and OGBench-Cube, particularly in high-novelty scenarios. The method enhances novelty-related structure in the planning latent and reduces multi-step prediction error.
Evidence and limits
Published automatically after robotics and source-evidence checks; no manual editorial approval is recorded. Source assertions are not independently verified. Missing information remains not reported.
- Environment:
- Not reported
- Control:
- Not reported
- Data origin:
- Not reported
- dataset: Not reported
Source excerpts and review record
No manual editorial approval recorded.
Original source quotation: “Code is available at https://anonymous.4open.science/r/atlas-world-model-72C4/”
Source E1
Source：arXiv Robotics — research abstracts · arxiv.org