This dataset contains 25,656 samples for training and evaluating ambiguity resolution capabilities in robot navigation and interaction systems. It covers common object reference ambiguities in indoor scenes.
Why This Dataset
Real human instructions are often vague, incomplete, or inconsistent with the environment, but existing VLN datasets assume perfect clarity.
This dataset introduces realistic ambiguity so that models… See the full description on the dataset page: https://huggingface.co/datasets/qiqiquq/AmbiguityDataset.
