72 hours of real robot operations, independently reconstructed
We ran our engine on MoRoOp, a public dataset from KTH Royal Institute of Technology: one autonomous mobile robot in a laboratory, nine eight-hour shifts. It is not a production fleet and not a benchmark of the robot.
Where the time went
Vendor 6.7 min
- Productive
- 55.4 h
- Planned charging
- 2.8 h
- Facility
- 10.9 h
- Unattributed
- 2.8 h
- Vendor
- 6.7 min
Real robot data, mapped to VDA 5050. Human-reviewed laboratory rules.
The robot itself caused about seven minutes of lost time.
Most lost time was idle with no work pending, which we assign to the facility.
2.8 hours remain unexplained by the data, and we show them as such.
Checked against the source
Our timeline matches the operation sequence the dataset’s authors published, within one second.
Our record counts match theirs exactly, 45 of 45.
Assumptions
Where the data could not decide, a person did, and we wrote it down. Time at a station during an active operation counts as process time. Pauses under ten seconds while driving count as driving. Scheduled charging is planned, not lost. Everything else we could not prove stays unattributed.
Source
Data: Klein, J.-F., Jeong, Y., “Autonomous mobile robot operations in production logistics: a dataset of jobs, dispatch events and robot states”, arXiv:2610.02428. Dataset DOI 10.71775/kth.vx0cq-w7h93, licensed CC BY 4.0. Mapped and analysed by Honest Machines; changes were made.