Papers/2609.13225
🧪 Test?View on arXiv

Part Grounding, Not Action Knowledge: Locating the Bottleneck in VLM Affordance Prediction

Author1, Author2, Author3, Author4, Author5

affordance predictionpart groundingmultimodalmodel evaluation
2609.13225
Builder Relevance
80%
2h ago

Abstract

This paper identifies part grounding as the primary bottleneck in vision-language models' ability to predict affordances for manipulation tasks.

Reality Card

Core Claim

Naming the target part significantly improves action accuracy and recall in affordance prediction tasks across multiple models.

Method / Result

Naming the target part raised action accuracy by 0.32 to 0.63 across all models tested.

Limitations

The study documented measurement errors that could affect the reproducibility of results.

Paper to code

Verified implementation resources so builders can test the paper’s claims instead of stopping at the abstract.

No verified implementation link has been attached yet. AIBuzzHub will keep this panel separate from unverified search results.
← Back to all papers