← All datasets

Warehouse-style object-manipulation dataset introduced with the spatiotemporal convolutional ergonomic-risk work (IEEE RA-L, 2019): 20 participants pick and place boxes and rods of varying weights at multiple heights in ~3-minute Kinect recordings with skeletal tracking. Every frame is annotated with one of 17 action labels on a four-tier hierarchy (object, motion, manipulation, surface height), of which 3 classes are low, 11 medium, and 3 high ergonomic risk - ground truth for the UW action-segmentation and REBA-prediction models.

Details

Subjects

n = 20 · 18-25

Protocol / load

Boxes and rods of varying weights picked and placed at multiple shelf heights

Equipment

Kinect for Xbox One (RGB video at ~12 fps plus Kinect skeletal tracking)

Sampling

video 12 Hz

License

CC-BY-4.0

Keywords

action-recognition, ergonomic-risk, object-manipulation, warehouse, kinect

Added to catalog

2026-08-24

Cite

Cite the dataset as its authors request (check the source page for an official citation); this BibTeX is a convenience starting point.

@misc{uw-iom,
  author = {Parsa, B. and Samani, E.U. and Hendrix, R. and Devine, C. and Singh, S.M. and Devasia, S. and Banerjee, A.G.},
  title  = {{UW-IOM: University of Washington Indoor Object Manipulation Dataset}},
  year   = {2019},
  doi    = {10.17632/xwzzkxtf9s.1},
  url    = {https://data.mendeley.com/datasets/xwzzkxtf9s/1},
}