corvinus-labs
regen_s2_200_seed29700
200 successful tube-rack insertion episodes from Isaac Sim with a myCobot 320 M5, using full appearance domain randomization (S2 config of staged regen). S2 maintains fixed geometry and single rack/tube model while randomizing table material, rack properties, tube appearance, liquid presence, and lighting.
Downloads20
Episodes200
Why This Matters for Physical AI
This dataset enables quantification of appearance domain randomization's impact on sim-to-real transfer for robotic manipulation by isolating appearance variation from geometric and environmental factors.
Technical Profile
- Modalities
- rgb
- Robot Embodiments
- myCobot 320 M5
- Action Space
- end_effector_delta
- Environment
- simulation
- Task Types
- manipulationinsertionpick_and_place
- Episodes
- 200
- Data Format
- PNG images with JSONL metadata
- Annotation Types
- action_labelslanguage_instructionsperturbation_labels
Access
Need custom rgb data?
Claru builds purpose-built datasets for simulation applications with dense human annotations and quality assurance.
Request a Sample Pack