Benchmark RadarAFFORDANCE20Q
Affordance20Q is a benchmark for evaluating affordance reasoning in LLMs using a 20-questions game. It comprises 1,009 games over 454 objects and 59 affordances, where models identify a hidden object's affordance by asking yes/no questions about physical properties.
- Released
- 2026-06-12
- Readiness
- Runnable
- Primary field
- General AI
Why it matters
Affordance reasoning is fundamental to physical understanding. Affordance20Q tests whether models can reason over physical properties without relying on memorized object-affordance mappings, which is crucial for embodied AI.
Motivation
Affordance reasoning, the inference of an object's action possibilities from its physical properties (e.g., shape and material), is fundamental to human physical understanding and increasingly critical for Large Language Models (LLMs).
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.