Dataset · OpenMOSS-Team
RoboOmni omnimodal robot actions
OpenMOSS-Team/OmniAction
Data for proactive robot manipulation in an omnimodal context.
What's inside
The dataset is related to the RoboOmni paper (ICLR 2026) on proactive robotic manipulation in an omnimodal context. The card notes that modern VLA models mostly rely on explicit instructions, whereas humans rarely give them in real interaction. According to metadata, the modalities are any-to-any and audio-to-audio, the language is English, and the size is not specified.
What you can do with it
- Training a VLA model for robotic manipulation
- Testing robot reaction to implicit instructions
- Exploring audio-to-audio scenarios in robotics
- Tasks
- Robotics, Multimodal, Audio → audio
- Languages
- English
- Downloads this month
- 89.5K
- Likes
- 296
How to load it
from datasets import load_dataset
ds = load_dataset("OpenMOSS-Team/OmniAction")