Language-guided 3D affordance grounding asks models to locate functional regions on 3D objects based on interaction intents. A new arXiv paper argues that current approaches generalize poorly to unseen objects, and it identifies the 'query-target' gap as one of the challenges behind this failure.

To address this, the authors propose BridgeAfford, a method aimed at bridging that gap. The paper, listed as arXiv:2610.21918, does not include further details of the architecture or experimental results in its abstract, so the specific mechanism remains unclear from the available text.

The abstract frames BridgeAfford as a response to the generalization problem, but does not report benchmark outcomes.