DynamicVLA Simulation Task Examples
A curated subset of 18 demonstration episodes from the DOM simulation environment spanning 3 task families (long-horizon, pick, place) with 207,306 total episodes and 140,932 structured instruction IDs. Data includes synchronized and individual camera feeds (opposite, wrist, side) with official action and state annotations in Parquet format.
Why This Matters for Physical AI
This dataset provides structured multi-view simulation demonstrations with language instructions for training vision-language-action models on complex long-horizon manipulation tasks with precise success criteria, enabling development of embodied AI systems that understand and execute natural language commands.
Technical Profile
- Modalities
- rgbproprioception
- Robot Embodiments
- Franka
- Environment
- simulation
- Task Types
- long_horizonpick_and_placepickingplacing
- Episodes
- 207306
- Data Format
- Parquet
- Annotation Types
- language_instructionsaction_labels
- License
- slab-license
Access
Need custom rgb data?
Claru builds purpose-built datasets for simulation applications with dense human annotations and quality assurance.
Request a Sample Pack