Skip to content

Dataset · acvlab

30k video episodes with actions and 3D annotations

acvlab/ABot-World-Explorer-500h

Video dataset tied to keystroke logs, text descriptions, and sparse 3D models.

What's inside

The dataset includes 30,969 video episodes (185,814 source objects total) associated with the ABot-World-0 infrastructure. Each episode contains MP4 video, keystroke logs, text captions, and a sparse COLMAP 3D text model. Materials are distributed under an open Apache-2.0 license.

What you can do with it

  • Training world models that predict environmental reaction to agent actions
  • Training multimodal navigation and control agents in virtual worlds
Size
10K–100K rows
Downloads this month
28.4K
Likes
30

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("acvlab/ABot-World-Explorer-500h")