Object goal navigation (ObjectNav) is a task in embodied AI where an agent must locate a target object, such as a bed, in an environment it has never seen before. The agent relies on spatial commonsense, including knowledge about object co-occurrence, to guide its search. The new paper from arXiv introduces CovNav, a method that combines large language models (LLMs) with conformal prediction to make inference adaptive.
The abstract, which is truncated, indicates that LLMs are involved in the approach. The key contribution appears to be the use of conformal prediction to adjust inference on the fly, though details of the mechanism are not provided in the available text. Because the source is a single arXiv abstract, the description here is limited to what is explicitly stated.
CovNav's focus on adaptive inference suggests an attempt to handle uncertainty in object navigation, but the paper's full methodology and results are not yet accessible from this announcement. Readers interested in the specifics should consult the full arXiv listing.