VLM Affordance Inference
Assessing VLM-driven semantic affordance inference for non-humanoid robot morphologies (AAMAS 2026).
This work, published at AAMAS 2026, assesses the zero-shot affordance inference capabilities of Vision-Language Models (VLMs) specifically for non-humanoid robots.
Key contributions:
- Demonstrates the limitations and capabilities of large foundation models to infer affordances in the context of non-humanoid robotics.
- Identifies an inherent conservative bias in model inference — an important observation for systems design and operational safety.
DOI: 10.65109/WTKR8312 arXiv pre-print arXiv:2604.19509