The legend. Reading the drawing.
Plain-English definitions for the automation and robotics terms that turn up across The Log — written from the deployment seat, as I relearn them myself. No prior knowledge assumed. Public sources only.
Throughputcases / hourHow many items the system moves in an hour. The number everything else is judged against.
Every spec, every payback math, every “up to” claim eventually resolves to this one number under real conditions. The demo number and the Tuesday-night-peak number are rarely the same.
Pick ratesuccess / attemptOf the times the robot tries to grab something, how often it actually gets it.
Easy to hit on clean, uniform, face-up items. The rate that matters is the one on whatever the trailer hands you — the crushed box, the creased label, the bag that wasn’t a box.
Exception handlingWhat happens when a pick fails, a label won’t read, or a tote jams — and who deals with it.
In an automated system the robots aren’t the operation; the exception handling is. When something goes wrong, what happens next is what decides whether the whole line is fast or slow. Humans didn’t leave — they moved to the edge cases.
SingulationSeparating a pile of items into a neat one-at-a-time stream the system can actually handle.
Robots and scanners want items presented one at a time, spaced out, the right way up. Most of the world arrives jumbled. A lot of “automation” is really just the unglamorous work of turning a heap into a line.
Peakpeak seasonThe busiest stretch of the year — the stress test no demo ever runs.
A pilot runs on forgiving timelines and attentive engineers. Peak runs on mixed freight, full volume, and nobody from the vendor in the building. Systems are bought on the demo and survive on peak.
End-of-arm toolingEOAT · the handThe business end of a robot arm — the part that does the grabbing.
Swap the hand and you change what the robot can do far more than swapping the arm. Most of the hard problem in picking lives in these last few inches.
Suction grippervacuumGrabs with a vacuum cup instead of fingers. The default for parcel picking.
Suction is fast, simple, and forgiving about exactly where it grabs. The catch: it hates anything porous, wrinkled, or wet — which describes a lot of real packaging.
Finger gripperparallel / mechanicalGrabs by closing on an object like a hand or pincer.
More versatile than suction, but it has to know the object’s shape and how hard to squeeze. Get either wrong and you either drop it or crush it. Knowing both, reliably, is the hard part.
Deformable itembag / poly mailerAnything that changes shape when you touch it — the opposite of a rigid box.
A rigid box is a robot’s best friend: known shape, flat faces, predictable grasp, scannable surface. A poly mailer has no stable grasp point, creases its own label, and shifts as the contents move. Retail keeps drifting from boxes toward bags. The robots are not thrilled.
Presentationpick faceHow an item is offered up to the robot — angle, spacing, which way it’s facing.
Demo inputs are clean, uniform, and face-up. Production inputs are whatever showed up. The machine doesn’t change between the two — the world feeding it does. That gap is most of the job.
ROSRobot Operating SystemNot an OS like Windows. The plumbing that lets a robot’s parts talk to each other.
It’s the shared language and wiring — sensors, motors, planners — passing messages so the pieces act like one machine instead of a box of parts. Most modern robotics work is built on top of it rather than from scratch.
SimulationsimTraining and testing a robot in a virtual copy of the world before it touches anything real.
A robot can run millions of virtual attempts overnight — cheap, fast, and nothing gets broken. The question is always how much of that virtual skill survives contact with reality.
Sim-to-real gapThe drop in performance when a robot trained in simulation meets the actual world.
Simulated friction, lighting, and floppy-physics never quite match reality. The robot that’s perfect in sim meets a real crumpled bag and discovers the world is weirder than its training. Closing that gap is much of the modern game.
Computer visionHow a robot “sees” — finding the item, deciding where to grab, reading the label.
Good light and clean items make it look easy. Glare, overlap, a torn label, two things touching — that’s where seeing turns into guessing, and guessing turns into an exception.
No-read rateThe share of labels the scanner can’t read on the first pass.
A couple of percent sounds harmless until you multiply it by everything moving downstream. Each no-read becomes a diverted item, a manual fix, a small jam in a fast line. The cost lives downstream, not where it happened.
Robotic armmanipulatorThe fixed, jointed arm that picks things up and puts them down.
The star of every demo video. Bolted in place, fed by a conveyor, judged entirely on what it does with whatever arrives at its reach.
AMRautonomous mobile robotA robot that drives itself around the floor, sensing obstacles and rerouting.
Moves inventory or totes to where the work is. It maps the space and adapts when something’s in the way — which is what separates it from its older cousin below.
AGVautomated guided vehicleThe older cousin of the AMR — follows fixed paths like wires or magnetic tape.
Simple and predictable, but it can’t improvise. Move the obstacle into its lane and it stops and waits. Cheaper to run, more expensive to change.
Trailer-unloading robotA machine that pulls freight out of a truck and feeds it into the building.
One of the harder jobs in the building: mixed freight, no presentation, gravity working against you. The quoted cases-per-hour range tends to be wide — and the freight is what decides where you land in it.
“Up to”Marketing’s favorite two words. Always the best case, never the floor.
“Up to 1,500/hour” answers the ceiling. The deployment lives at the floor. Ask what the worst sustained hour looks like, not the best demo minute.
UptimeThe share of time the system is supposedly running.
“99% uptime” — measured over what window, and excluding what? Scheduled downtime, jams cleared by a human, the holiday week — the asterisks are where the truth hides.
The rangee.g. 400–1,500/hrWhen a spec gives a wide spread instead of one number, the spread is the story.
A 4x range isn’t a footnote — it’s the difference between a payback-in-18-months machine and a very expensive decoration. What lands you at the bottom of the range is almost always the freight.
Payback periodROIHow long until the machine has saved more than it cost.
Every payback model quietly assumes the top of the range. Re-run it at the bottom — the freight you’ll actually feed it — and see if the story still holds.
Pilot vs. deploymentA controlled trial versus the thing running for real, every shift.
A pilot runs on a curated slice of reality: clean inputs, attentive engineers, forgiving timelines. A deployment runs on peak, mixed freight, and a Tuesday night with nobody from the vendor in the building. Same robot. Different planet.