Dataset · yssssjtu
Precomputed caches for robot training moqi_data
yssssjtu/moqi_data
Tarballs with MobActWorld-H3 caches based on AgiBotWorld2025, Galaxea, and BEHAVIOR-1K demonstrations.
What's inside
A set of precomputed MobActWorld-H3 training caches collected from robotics datasets AgiBotWorld2025, Galaxea, RoboChallenge v2, and BEHAVIOR-1K. The data is split into 10 batches as uncompressed tar archives with 73-frame window metadata and JSONL files with SHA256 checksums.
What you can do with it
- Accelerated training of video-to-video generation and action models for robotics
- Training multi-dataset manipulator control policies
- Reproducing experiments based on standardized 73-frame windows
- Tasks
- Video → video, Robotics
- Downloads this month
- 10K
- Likes
- 0
How to load it
from datasets import load_dataset
ds = load_dataset("yssssjtu/moqi_data")