Industries / Health / Healthcare / Medical labs
Collecting and moving samples
Drawing blood and other samples, labeling each tube, and getting it to the analyzer in the state it left the patient.
6
Milestones achieved, from 1949 to 2024.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1949
A tube that fills itself
Blood used to be drawn into a syringe and poured into test tubes, which often spilled, and each test needed its own needle stick. Joseph Kleiner, a mechanical engineer, built a sealed tube holding a vacuum that pulls blood from a vein straight into a sterile tube, which can also hold chemical additives. By BD's account the patent was filed in 1947 and the first BD Vacutainer tubes went on sale in 1949; BD says 84 people every second now have blood drawn with its Vacutainer products.
Credit
Joseph Kleiner; Becton, Dickinson and Company (BD)
1981
A lab on conveyor belts
Kochi Medical School in Japan opened in 1981 with only 19 laboratory technicians allowed for its first three years, so its lab staff built their own system of conveyors and robots to carry tubes between analyzers. About ten years later such systems went on sale in Japan, and by the team's 1998 count 72% of Japan's national university hospitals had installed one. The team wrote that the systems were still too costly for small hospitals and asked makers to agree on common rack shapes and to use more flexible robots.
Credit
M. Sasaki and colleagues, Kochi Medical School
1997
Most errors precede the test
At the University Hospital of Padova, Italy, two researchers tracked 40,490 urgent test results over three months and found 189 mistakes, or 0.47%. Of these, 68.2% happened before the analysis (in ordering, collection, labeling or transport), 13.3% during it and 18.5% after it. In 12 patients the mistakes led to wrong care or a wrong change of treatment. A 2026 review puts this "preanalytical" share of lab mistakes at 68% to 98%.
LessonA more precise analyzer does not fix most lab errors, which happen in collection, labeling and transport.
Credit
M. Plebani and P. Carraro
2005
A tube coating skews results
After some immunoassays (tests that use antibodies to catch a hormone or protein) gave wrong results on blood held in BD's SST tubes, a team at the US National Institutes of Health Clinical Center traced the cause to Silwet L-720, a soap-like coating on the inside of the tubes. At 400 mg/L in serum it raised the measured total triiodothyronine, a thyroid hormone, by about 51% on one analyzer, the IMMULITE 2500, but not on another, the AxSYM, because it pulled the test's capture antibodies off their solid surface.
LessonThe tube is part of the test, and any change to its materials has to be checked against each assay.
Credit
Raffick A. R. Bowen and colleagues, NIH Clinical Center
2015
Drones carry lab samples
Researchers at Johns Hopkins University drew paired tubes from 56 volunteers and flew 168 of them on a small drone for 6 to 38 minutes; results for 33 common chemistry, blood count and clotting tests matched the tubes kept on the ground. A 2017 follow-up by the same lead author flew samples 258 km over 3 hours, and glucose and potassium shifted by 8% and 6.2% because the flown samples ran about 2.5 °C cooler. From March 2019, UPS Flight Forward flew blood and urine samples by Matternet M2 drone between two WakeMed sites in Raleigh, North Carolina, a quarter mile apart, about eight round trips a day, as CAP Today reported.
LessonLong flights need temperature control, or results drift.
Credit
Timothy K. Amukele and colleagues; WakeMed, UPS Flight Forward and Matternet
2024
Fingertip draws for routine panels
BD's MiniDraw device, used with Babson Diagnostics' BetterWay service, gives routine test results from as few as six drops of blood from a fingertip, collected by trained health workers who need not be phlebotomists (specialists in drawing blood from veins). It has been in use at US retail pharmacies since May 2024 and was offered to health systems from December 2024. A Babson study published in May 2025 found 37 of 39 tests matched blood drawn from a vein within set error limits; CAP Today noted that the company ran the study and that it does not cover patients collecting their own blood.
Credit
BD; Babson Diagnostics
Watch for
A robot draws blood
Vitestro's Aletta finds a vein with near-infrared light and ultrasound, then inserts the needle, changes tubes and applies a bandage without a person's hands; one phlebotomist supervises. In a company-backed multicenter trial published in Clinical Chemistry, with 1,633 patients at three outpatient blood-draw departments, Vitestro reports a first-stick success rate of 94.5% when the robot found a suitable vein, hemolysis (burst red cells that spoil some results) in 0.3% of samples and adverse events in 0.6%. The US Food and Drug Administration (FDA) granted De Novo authorization, its route for new kinds of devices, in August 2026. The robot is CE marked and in pre-commercial use in Europe, and Vitestro plans a US trial before selling it there.
Achieved whena hospital not funded by Vitestro publishes success, hemolysis and complication rates from routine use, including how often the robot declined to draw.
Claimed by
Vitestro, April and August 2026
Open
01
A robot that racks tubes
A low-cost robot arm that takes mixed blood tubes from a bin, reads each barcode, and places each tube in the rack slot for its destination analyzer, trained from human demonstrations. Use practice tubes with colored caps, filled with dyed water and labeled with made-up barcodes, so no real samples are involved.
The briefCredit · unclaimed02
A tube photo test set
A public set of photos of mock blood tubes with made-up patient labels, some correct and some with a known fault: no label, a partial or smudged label, a name that does not match the mock order, the wrong cap color for the ordered test, too little liquid, or a label covering the window used to check the fill level. Then score AI models that read images on finding each fault.
The briefCredit · unclaimed
Open 01 · the brief
A robot that racks tubes
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A low-cost robot arm that takes mixed blood tubes from a bin, reads each barcode, and places each tube in the rack slot for its destination analyzer, trained from human demonstrations. Use practice tubes with colored caps, filled with dyed water and labeled with made-up barcodes, so no real samples are involved.
Why it is still open
The Kochi team asked in 1998 for flexible robots that small labs could afford. Large sorters that route tubes by barcode exist, such as the robot cells the American Red Cross began using in 2007 to sort 2,700 tubes an hour, but they are fixed machines programmed by hand. The closest published work with a learning robot, from Fujitsu and Waseda University in March 2026, trained a two-armed robot to set chemistry test tubes in a rack, with an 86.6% average success across three tasks; it did not route clinical tubes by barcode. A hobby model on Hugging Face, trained in September 2026 to sort tubes by color, publishes no success count.
Built with what exists
an SO-101 arm and Hugging Face's LeRobot software (see the Note "No download"); a small vision-language-action model, one that turns camera images and an instruction into arm movements, such as SmolVLA, NVIDIA's GR00T 2 (see the Note "Five demonstrations per task") or Runway's open Praxis-1 weights (see the Note "Open robot model weights"); a USB camera; the open-source ZXing barcode reader. The advance is robot models that learn a pick-and-place task from tens of recorded demonstrations instead of hand-written motion code.
First step
Buy 60 practice tubes in six cap colors, fill them with dyed water, print barcodes for ten made-up destinations, and record 50 demonstrations with LeRobot.
Done looks like
300 filmed attempts with six cap colors and ten destinations, and a published tally of correct slot, wrong slot, dropped tube and stalled, with at least 95% placed correctly and no dropped tubes. The trained model and demonstration data are released, and a second team repeats 50 attempts on its own arm and comes within 5 percentage points.
Open 02 · the brief
A tube photo test set
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of photos of mock blood tubes with made-up patient labels, some correct and some with a known fault: no label, a partial or smudged label, a name that does not match the mock order, the wrong cap color for the ordered test, too little liquid, or a label covering the window used to check the fill level. Then score AI models that read images on finding each fault.
Why it is still open
A College of American Pathologists survey of 147 laboratories found 0.92 labeling errors per 1,000 labels across 3.3 million labels. In 2014 ARUP Laboratories built a four-camera station that read patient names by optical character recognition (OCR) and caught 121 mislabeled tubes in about a million images, 50 of which its usual checks had missed, but it compared names only. A 2026 review finds that commercial systems still rely on barcodes, sensors and fixed rules, and a search found no public image set of tube faults.
pubmed.ncbi.nlm.nih.govpubmed.ncbi.nlm.nih.govpubmed.ncbi.nlm.nih.gov
Built with what exists
practice tubes, dyed water and a label printer; a phone camera; AI models that read images, such as Claude, Gemini and GPT, and open-weight vision models such as Qwen2.5-VL; Tesseract OCR as a plain baseline; a scoring script. The advance is AI models that read a photo and an order form together and say what does not match.
First step
Print labels for ten made-up patients, make 50 tubes, 15 of them faulty, and test one model.
Done looks like
2,000 photos of 500 tubes, 30% of them with one fault, keyed by two people who agree on every photo after review. At least five models and the OCR baseline are scored on the share of each fault type caught while flagging no more than 2% of good tubes, and photos, key and code are public under an open license.