Mobile manipulators face a difficult trade-off: they need to keep several plausible ways to perform a task while still reacting quickly during execution. The arXiv preprint "MobiGuide: Vocabulary-Guided Online Recovery for Mobile Manipulation" frames this as a central challenge in the field. It offers the example of grasping a bottle in a cluttered scene, where multiple physically valid grasps are possible.

The paper proposes a method it calls vocabulary-guided online recovery, as indicated by the title and abstract. The idea is that a structured vocabulary can guide the system toward an alternative action mode during recovery, though the abstract does not include implementation details or experimental results.

This digest is based on a single source, so there are no independent or conflicting findings to compare.