tommypraxis
Praxis Egocentric Evaluation Samples
Real-world hand, object and tool interaction dataset across six camera views with MCAP sensor data, MP4 previews, and camera calibration. Includes 735 samples spanning 24.16 hours of mapped activity across 26 primary task categories and 15 environment types.
Downloads17
Episodes735
Hours24.16
Why This Matters for Physical AI
This egocentric dataset of real-world hand-object interactions with precise camera calibration provides essential multi-view visual data for training embodied AI systems to perform diverse manipulation and interaction tasks in uncontrolled real-world environments.
Technical Profile
- Modalities
- rgbcamera_intrinsicscamera_extrinsics
- Environment
- commercial_kitchenconstruction_sitefactoryfarmhospitality_venuelaboratorymineofficeoutdoor_groundsresidentialretailwarehouseworkshop
- Task Types
- animal_careassemblycleaningconstructioncookingcuttingdisassemblyequipment_operationfoldingfood_preparationgardeninginspectionlabelingmaintenance_repairmaterial_transportmeasuringpackingpaintingscanningservingsewingsortingunpackingwashing
- Episodes
- 735
- Total Hours
- 24.16
- Data Format
- MCAP
- Annotation Types
- activity_labelsenvironment_labelsdescriptive_titlescaptionssecondary_activities
Access
Need custom rgb data?
Claru builds purpose-built datasets for commercial_kitchen applications with dense human annotations and quality assurance.
Request a Sample Pack