COAT

Useful for checking the physical reasoning capabilities in household agents. Made through human annotations of what type of object configurations we prefer for accomplishing a household task.

Consists of 4 major datasets that are organized for evaluating such object selection capabilities across 3 major factors 1) Utility [1 dataset] 2) Context [1 dataset] 3) Physical State [2 datasets]

We also provide 2 datasets used for evaluating object selection capabilities by applying all 3 factors together. Namely the F_{ideal} and F_{moderate} datasets.