ADAPT: Benchmarking Commonsense Planning under Unspecified Affordance Constraints

Pei-An Chen, Yong-Ching Liang, Jia-Fong Yeh, Hung-Ting Su, Yi-Ting Chen, Min Sun, Winston H. Hsu · 2026

Read ADAPT: Benchmarking Commonsense Planning under Unspecified Affordance Constraints by Pei-An Chen, Yong-Ching Liang, Jia-Fong Yeh, Hung-Ting Su, Yi-Ting Chen, Min Sun, Winston H. Hsu (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-14902.txt.

Description

Chen and six others introduce DynAfford, a benchmark where embodied-agent affordances shift over time unstated in instructions. Agents must perceive object states, infer hidden preconditions, and adapt actions to match. They add ADAPT, a plug-and-play module giving existing planners explicit affordance reasoning. ADAPT lifts robustness and task success across seen and unseen settings. A domain-adapted, LoRA-finetuned vision-language model as backend even beats GPT-4o, showing the value of task-aligned affordance grounding.

Catalog record

CREATORPei-An Chen, Yong-Ching Liang, Jia-Fong Yeh, Hung-Ting Su, Yi-Ting Chen, Min Sun, Winston H. Hsu
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE55639
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.14902

Browse the full library · About the Tower of Babel