Skip to content

Dataset · spatialverse

2 million pairs for Vision-Language Navigation

spatialverse/SAGE-3D_VLN_Data

2 million trajectory-instruction pairs across 1,000 indoor environments for training vision-language navigation.

What's inside

A Vision-Language Navigation dataset containing 2 million trajectory-instruction pairs from 1,000 indoor scenes. Instructions are hierarchical, ranging from high-level semantic goals to low-level control commands. Covers two main task types—VLN and No-goal. Format and language are not specified in the dataset card.

What you can do with it

  • Training a model or policy for VLN using hierarchical instructions
  • Evaluating a robotic agent across 1,000 indoor scenes
  • Performing SFT for image-text-to-text navigation
Tasks
Robotics, Image text → text
Downloads this month
934
Likes
69

How to load it

from datasets import load_dataset

ds = load_dataset("spatialverse/SAGE-3D_VLN_Data")