A picker is a camera, a stage and a set of pins or a tip head, and it does one thing: find colonies on a plate against criteria you set, and transfer them into an ordered destination. That is worth automating for two reasons rather than one. The obvious one is throughput, and the number of plates at which it beats a person is higher than vendors imply. The less obvious one is the record: a machine picks by measured criteria and writes down which colony went to which well, which is exactly what a person with a marker pen does not do reliably at eleven at night.
Colony picking robots, and what picking accuracy means
Accuracy here is two numbers that get conflated. The first is whether the machine picked a colony at all rather than agar, which modern imaging does well on a clean plate and badly on a crowded or a satellite ridden one. The second is whether it picked the colony it recorded, which depends on the stage repeatability and on whether the plate moved. Ask for both against a plate like yours, and ask what the failure looks like: a missed pick that is recorded as empty is a minor nuisance, and a pick recorded in the wrong well quietly corrupts a whole screen.
The imaging decides which colonies you can select on
White light and a size threshold is enough to pick the biggest colonies. Selecting on colour, on fluorescence, on a blue and white screen or on morphology needs the right illumination and filters, and those are options rather than defaults. This is the specification to get right before the mechanics, because it determines which experiments the machine can run: a picker that cannot see your reporter is a machine for picking anything at all, which is rarely the job.
Pins against tips, and what each costs to run
Steel pins are sterilised between picks by flame or by wash and last for years, so the running cost is nearly nothing and the risk is carry over if the cycle is poor. Disposable tips remove carry over entirely and add a consumable per pick, which at screening scale is the dominant cost. The choice follows the work: a library where a single cross contaminated well ruins a hit list wants tips, and routine picking of transformants is fine on pins with a validated wash.
The plate counts at which it pays
A person picks a few hundred colonies in an hour and gets slower and less accurate after that. So the machine wins clearly above a few thousand picks a week, or wherever the picking has to happen on a schedule nobody wants to keep. Below that, it competes with a person only on the record rather than on the time, and the honest comparison includes the plate preparation the machine needs (consistent pour, consistent spread, colonies in its size window), because a badly prepared plate costs more time at the machine than it saves.
Where it sits beside the rest of the bench
A picker produces ordered plates of cultures, which is the input to a preparation step and then to sequencing or screening. The integration worth having is not a robotic arm but a file: the machine's map of colony to well, in a format the downstream system reads. Where that file has to be retyped, the automation has moved the manual step rather than removed it, so ask about the export before the arm.
What to test during an evaluation
Bring your own plates, including a bad one. Run a full destination plate, then check the map against what grew: every well occupied, every culture the colony it claims, and the map exported into whatever comes next. Time the whole cycle including the loading and the sterilisation rather than the picking rate, and ask who cleans it and how long that takes at the end of a run, because that is the part a demonstration never shows.
Questions people ask about colony picking robot
Can it pick from a crowded plate?
Less well than a person can, and that is the limit that matters. The machine needs separated colonies in a size window, so the plate preparation upstream (dilution, spread, incubation time) is part of the automation rather than a detail. Plan to dilute more than you would by hand.
Does it replace a person?
It replaces the picking, not the plates. Somebody still pours, spreads, incubates, loads and cleans, and the machine's throughput is only reachable if that pipeline keeps up. The saving is real and it is smaller than the picking rate suggests.
Pins or tips for a library screen?
Tips, unless the wash cycle has been validated on your organism. One cross contaminated well in a hit list costs more than a year of tips, and the failure is invisible until somebody tries to reproduce the hit.