Datasets
Example validation results
CARLA-COCO
CARLA · 5,000 scenes
51
Overall
78
Coverage
0
Physics
80
Dist.
51
S2R
SynTable
Isaac Sim · 3,200 scenes
58
Overall
85
Coverage
0
Physics
95
Dist.
58
S2R
Warehouse Robotics
Isaac Sim · 1,240 scenes
63
Overall
74
Coverage
42
Physics
72
Dist.
60
S2R
Deep Dive
Warehouse Robotics — Full Report
63
Overall
74
Coverage
42
Physics
72
Distribution
60
Sim-to-Real
Coverage Breakdown
Object Distribution
7438/52 object classes represented
Spatial Coverage
7176% of navigable area covered
Viewpoint Diversity
688 camera angles, 2 heights
Scale Diversity
650.5x to 3x scale range
Physics Violations
| ID | Description | Severity | Scenes |
|---|---|---|---|
| PV-001 | Floating objects detected in 12 scenes | high | 12 |
| PV-002 | Interpenetrating meshes in shelf region | medium | 18 |
| PV-003 | Unrealistic shadow angles (light source mismatch) | medium | 24 |
| PV-004 | Objects clipping through conveyor belt | high | 8 |
Recommendations
- Add 14 missing object classes: fire extinguisher, safety cone, pallet wrap, label printer, hand scanner, forklift variant B, caution sign, dock plate, shrink wrap, barcode reader, safety goggles, hard hat, high-vis vest, floor marking
- Increase low-light scene variants (currently 4% of dataset, recommend 15-20%)
- Fix floating object physics in warehouse aisle scenes — 12 scenes affected
- Add forklift motion blur variants for dynamic scene coverage
- Expand viewpoint diversity: add overhead and floor-level camera angles
- Increase spatial coverage in loading dock and cold storage areas
Score Over Time
Track improvement across dataset versions
Warehouse Robotics validation scores improved from 38 to 63 over five iterations by following Lucitra's recommendations.
Validate your own datasets
Upload a dataset and get your first validation report in minutes.