Skip to content

Dataset · astralhf

Indoor drone navigation dataset and benchmark

astralhf/yonder

The largest open dataset with 4.65 million drone frames and a benchmark for testing model transfer between simulators.

What's inside

The dataset contains 4.65 million first-person drone frames for indoor navigation and accompanies a NeurIPS 2026 publication. In addition to the data itself for object detection, depth estimation, and robotics tasks in English, the project includes a closed benchmark to check perception degradation when transferring models from one simulator to another.

What you can do with it

  • Training visual drone navigation and indoor depth estimation systems
  • Testing the robustness of robotic models during transfer between simulators
Tasks
Object detection, Depth estimation, Robotics
Languages
English
Size
up to 1K rows
Downloads this month
20K
Likes
0

How to load it

from datasets import load_dataset

ds = load_dataset("astralhf/yonder")