Warehouse & Material-Handling Training Data
What is warehouse and material-handling data for robots?
Warehouse and material-handling data is first-person video of workers picking, packing, palletising and moving goods across many SKUs and conditions - with pose and action labels. It gives robot policies the object and environment diversity that drives generalisation for logistics manipulation.
What nxted captures
Recorded across realistic warehouse settings and item types.
- ›Pick and place across many SKUs and shapes
- ›Packing, sorting and palletising
- ›Tote and conveyor handling
- ›Action segmentation, hand pose and success labels
- ›Lighting, clutter and layout variation
Why diversity beats volume here
Logistics generalisation depends on breadth of objects and environments more than raw hours - the finding of the ICLR 2025 imitation-learning scaling-laws paper. A diverse warehouse-handling set across many SKUs and layouts is high-value training data.
How nxted delivers it
Robotics-ready in LeRobot, RLDS and HDF5 with metadata, a dataset card and a Data Trust Pack. Begin with a Physical AI Test Kit.
FAQ
What does warehouse-handling data include?
First-person demonstrations of picking, packing, palletising and moving goods across many SKUs, with action and pose labels.
Why does object diversity matter?
Robot-policy generalisation tracks the number of distinct objects and environments seen, so a broad SKU and layout range outperforms a narrow, repetitive set.
What formats?
LeRobot, RLDS and HDF5 with metadata, dataset card and QA report.