Dataset from the paper: "The Role of Variability in Human Navigational Instructions in Visual-Language Robot Navigation"

This dataset accompanies "The Role of Variability in Human Navigational Instructions in Visual Language Robot Navigation." It contains spoken navigation instructions collected from 225 globally-recruited participants guiding four different robot embodiments (a baseline, Blue, Player 2, and Unitree G1) through Matterport3D indoor environments. Each instruction is paired with participant trait data (sense of direction, perspective taking, need for cognition, extroversion, age and robot perception ratings). The paper uses this data to show that state-of-the-art VLN models are highly sensitive to natural linguistic variability in human instructions, even when environments and trajectories are held fixed.