VLA-Arenaapache-2.0
VLA-Arena Dataset (L0 - Large Variant)
An open-source benchmark for systematic evaluation of Vision-Language-Action (VLA) models with 60 tasks at difficulty level 0, containing 3,000 human demonstrations (50 per task) across safety, distractor, extrapolation, and long horizon domains. The dataset is strictly formatted in RLDS with high-resolution RGB images (256x256), robot state vectors, 7-DoF continuous control signals, and natural language task instructions.
Downloads222
Technical Profile
- Modalities
- rgbproprioceptionlanguage
- Environment
- simulation
- Task Types
- manipulationgraspingpick_and_place
- Data Format
- RLDS
- License
- apache-2.0
Access
Need custom rgb data?
Claru builds purpose-built datasets for simulation applications with dense human annotations and quality assurance.
Request a Sample Pack