Real-time AI-driven quality control for laboratory automation: a novel computer vision solution for the opentrons OT-2 liquid handling robot

1. Motion—Motion accuracy: not reported
2. Volume—Volumetric accuracy: not reported
3. Viability—Cell viability: not reported
4. Unattended—Unattended operation: not reported
5. Standard—Standard compliance: not reported

The full text was read and states no figure for any of the five.

Each score is shown against the range the other 59 builds actually occupy, because no criterion uses the full 1–5 scale. These describe how reproducible the build is, not how good the science is.

• Resolution

  3/1–5

  Vision volume estimation validated only over 50-270 uL (3.4% mean error, 95% accuracy) with P20 tips imaged at 10/20 uL; sub-10 uL capability is the host OT-2 and unmeasured.

• Throughput

  5/2–5

  Host OT-2 8-channel arm imaged with all eight tips and mapped across a 12x8 96-well plate, a parallel multi-tip head, though channels are not independently controlled.

• Parts sourcing

  3/2–4

  Part sourcing complexity not explicit; assigned moderate default.

• Skill needed

  2/2–4

  Multiple advanced technical skills appear required.

• Facilities needed

  3/2–4

  Requires accessible but nontrivial lab/maker equipment.

• Documentation

  3/2–5

  Some open documentation or files are available.

• Validation effort

  3/2–4

  50 live runs, 100 volume tests, mAP@50 of 0.995, 98% missing-tip accuracy and regression calibration, but Table 5 compares only to a prior YOLOv5 paper, not a conventional or manual benchmark.

• Cycle time

  3/1–4

  Only 1.1 s/frame versus 8 s for the prior AI method; no timing comparison reported against manual pipetting or a commercial handler, and vision adds per-step overhead.