DevanshL2026MIT
Training-free amodal grasping, experiment data
Experiment data and results for training-free primitive completion of occluded objects using amodal reasoning for robotic grasping. Includes LIBERO-Occ and YCB-Video benchmark scenes with ground truth, masks, VLM replies, method fits, and robot pick episodes.
Downloads57
Why This Matters for Physical AI
This dataset enables research in amodal object completion and occluded object manipulation, which is critical for robust robotic grasping in real-world scenarios where objects are partially hidden.
Technical Profile
- Modalities
- rgbdepthsegmentation
- Environment
- simulationlab
- Task Types
- graspingamodal-completion
- Annotation Types
- segmentationbounding_boxeslanguage_instructions
- License
- MIT
Access
Need custom rgb data?
Claru builds purpose-built datasets for simulation applications with dense human annotations and quality assurance.
Request a Sample Pack