Worked budget for a 10,000-hour teleoperation dataset with language annotation for a VLA model (DexSet)
~$646,000, or $64.60 per usable annotated hour per program (10,000 usable hours)
Dated 2 Jul 2026Source checked 11 Oct 2026Quote found on its page again 17 Oct 2026Basis: vendor
In the source's own words
“Total: ~$646,000, or $64.60 per usable annotated hour”
Where it was read
Also read from the same source
- Annotation passes on robot data, added to any capture modality (DexSet benchmark)
- Dense object masks or contact annotation on robot data (DexSet per-pass price)
- Collector labour inside egocentric video capture (DexSet cost structure)
- Language-instruction labelling of robot data (DexSet per-pass price)
- Multi-view exocentric capture for robot learning (DexSet benchmark)
- Operator labour inside teleoperated bimanual data (DexSet cost structure)
What it is about
Figures it counts in
- $38 per hour of videoPhysical-world data, made to order, per hour of video
- 2.7xHow much more buyers pay than makers are paid for physical-world data, per hour of video