Warehouse Inventory Glossary
The working vocabulary of warehouse inventory, defined. Operations, finance and technology teams all use these terms, and not always the same way. Each entry stands on its own, so you can read one and leave.
Category and approach
Embodied AI
Embodied AI is a form of physical AI where the intelligence comes from having a body and learning through moving around an environment. The system builds its own understanding of a space using onboard cameras and computer vision, then uses that understanding to navigate and act. Every embodied AI system is physical AI. The reverse does not hold, because a fixed camera running inference over a conveyor acts in the physical world without ever having to find its way through one.
Corvus One® is embodied in the strict sense. It flies warehouse aisles it mapped itself, reading the building as it goes, with no GPS and nothing added to the racking or the ceiling.
GPS-denied
A GPS-denied environment is any space where satellite positioning does not reliably work. Inside a warehouse the roof and the steel racking can block the signal, so a machine that relies on GPS to know where it is has nothing to rely on. This is a large part of why outdoor drone technology does not transfer well indoors.
A machine that cannot locate itself cannot report where anything else is. Corvus One and Corvus Trident™ both localize from computer vision, reading the building itself, so neither needs GPS, beacons, or markers.
Infrastructure-free
Infrastructure-free describes a system that works in your building without anything added to the building. No ceiling beacons, no rack markers, no navigation stickers, no reflectors, no dedicated Wi-Fi, no special barcodes.
Infrastructure-free covers both setup and upkeep. An infrastructure-free system maps the space itself and works from that map. No hardware has to be installed and surveyed in first, and none has to be maintained afterward to keep the system accurate. Corvus One needs one Ethernet connection and one power drop for its landing pad, and that is the extent of what the building provides.
Physical AI
Physical AI is artificial intelligence that operates in the physical world rather than only inside a data center. It runs on machines that perceive their surroundings through onboard sensors, reason about what they find, and act in real time. The action is physical even when the output is data, because the machine still has to get itself to the right place, pointed the right way, in a building that changes around it. A language model can answer a question about a warehouse. Physical AI has to go into the warehouse.
The distinction matters for one practical reason. Software that analyzes inventory data can only work with what somebody already counted. Physical AI produces the data itself by going and looking. Corvus One and Corvus Trident are both physical AI systems, with the cameras, the compute, and the decisions all running on the device.
Physical economy
The physical economy is the part of the economy that moves actual goods. Warehouses, distribution centers, manufacturing plants, freight, and the inventory sitting in all of them. It is the counterpart to the digital economy, and it runs on records that are only as good as the last time somebody checked them.
The gap between the digital record and the physical reality is where the work is. A WMS holds a number. The pallet either matches it or it does not. Autonomous inventory intelligence for the physical economy means closing that gap inside the building, as work happens.
RTLS (real-time location system)
RTLS stands for real-time location system, meaning any system that tracks where people, vehicles, or inventory are inside a facility as they move, instead of recording where they were at the last scan. Traditional indoor RTLS works by installing reference points around the building and measuring a tag against them, which is where ceiling beacons usually come from.
Corvus Trident is a vision-based RTLS with no reference points to install. It localizes the vehicle from what its cameras see, so the system knows where a forklift is and which pallet it just handled without a beacon network, a tag on every load, or a scan gun in anyone’s hand.
Localization
Ceiling beacon
A ceiling beacon is a battery-powered device mounted overhead that a robot uses to work out its position indoors, taking the role GPS plays outdoors. Beacons have to be installed and mapped before the system runs, maintained for the life of the deployment, and moved when the layout changes. See localization infrastructure.
Localization infrastructure
Localization infrastructure is the hardware a robot needs installed in your building so it can work out where it is. Ceiling beacons, rack markers, navigation stickers, retroreflective targets. Without these reference points the robot does not know where it is in the facility.
The hardware has to be installed and its positions recorded before the system can use it, and it has to stay where the system expects to find it. Battery-powered beacons have to be maintained and moved when the layout changes. Passive markers and stickers have to stay clean, visible, and in position.
Corvus One and Corvus Trident require none of it. Both work out their own position from what their cameras see. The facility still gets mapped, because any system needs to know the layout it is working in. Nothing gets installed on the racking or the ceiling to make that mapping possible.
Rack marker
A rack marker is a printed target applied to racking that a robot’s camera reads to work out where it is. It needs no power, unlike a beacon, but it does need to be placed accurately, mapped, kept clean and visible, and put back correctly whenever racking moves. Markers go on the racks. Beacons go on the ceiling. See localization infrastructure.
Counting and accuracy
Cycle counting
Cycle counting is the practice of counting a portion of inventory on a rolling schedule so the whole facility gets verified over time, instead of shutting down for one annual physical inventory. The common method is ABC classification, which counts fast-moving SKUs far more often than slow movers.
The method is sound. The constraint is labor. A team on foot with a scanner and a lift can only reach so many locations per shift, so counting frequency gets set by headcount rather than by risk. Automated cycle counting removes that ceiling. Corvus One flies the same aisles on any shift, including lights-out, which makes count frequency a scheduling question rather than a hiring question.
Inventory accuracy
Inventory accuracy is the percentage of inventory locations where what the system says matches what is physically there. It is measured by counting and comparing, so a facility only knows its accuracy as of its last count. A number quoted from an annual physical inventory describes the building on one day, months ago.
This is the metric most warehouse problems trace back to. Stockouts, overstock, mispicks, chargebacks, and write-offs are all downstream of a record that disagrees with the rack. It is also why counting frequency and accuracy are the same conversation. A facility counted once a year has an accurate number once a year, and an estimate for the other three hundred and sixty four days.
Inventory discrepancy
An inventory discrepancy is any case where the system record and the physical reality disagree. A pallet in the wrong location. A location the system thinks is empty that is not. A quantity that is short, or long, or a SKU that is somewhere nobody expected.
Discrepancies are the actual output of a count. The count itself is only the method. Detection speed matters more than detection volume, because a discrepancy found the night it happens can be traced to a shift and a movement, while one found eleven months later is an adjustment entry. Southern Glazer’s Wine & Spirits identified over 35,000 verified discrepancies across nine distribution centers.
Inventory visibility
Inventory visibility is knowing what inventory you have and where it is, at a given moment, across every location you run. It covers the count, the location, and how current both are. A facility with high accuracy and quarterly counts has poor visibility, because the information is right and old at the same time.
Inventory tracking is the same idea stated as an activity. Visibility is the state. Tracking is the work that produces it. Corvus One covers inventory at rest by flying the aisles. Corvus Trident covers inventory in motion by riding the forklifts and recording every pallet move. Together they keep the record current.
Physical inventory (PI)
A physical inventory, usually shortened to PI, is a full count of everything in a facility checked against what the system says should be there. It is the catchall that reconciles a year of drift, and it is traditionally done wall to wall, with the building shut down or running at reduced capacity while teams work through every location.
The annual PI exists because continuous counting was impractical. It is expensive, it is disruptive, and by the time it finishes the earliest counts are already stale. Autonomous counting changes the cadence rather than the concept. At MSI Surfaces, PI moved from an annual catchall to a weekly validation pass at each facility, which is the same reconciliation run often enough that drift never accumulates.
Shrink
Shrink is inventory a business owns on paper and cannot find in reality. Theft, damage, mislabeling, receiving errors, and product that was recorded and then lost track of. It gets written off, and the write-off lands in finance rather than in operations, which is part of why it persists.
Some of what gets called shrink is recoverable. A meaningful share is inventory that exists and is simply in the wrong place, invisible to a system that thinks it is elsewhere. Accurate, frequent counting separates the two. Product found in the wrong location is recovered instead of written off, and the pattern of where it keeps ending up tells you which process is producing it.
Operations
Lights-out
Lights-out describes a facility or a process that runs with nobody there. The term comes from manufacturing, where a lights-out shift is one the machines work alone, and it transfers directly to inventory counting.
Counting is unusually well suited to it. Counting wants the building quiet, ideally with nothing moving, which is exactly when a warehouse is least useful for anything else. Corvus One flies lights-out because nobody has to launch it, follow it, or swap its batteries, so a count can run overnight and be waiting when the first shift arrives. At LAPP USA, lights-out flights begin at 6 PM and the facility gets 26 full-warehouse counts a year.
Material handling equipment (MHE)
Material handling equipment, shortened to MHE, is the powered equipment that moves goods inside a facility. Forklifts, reach trucks, order pickers, pallet jacks, and the rest of the fleet that touches inventory between the dock and the rack.
MHE is a constraint as well as an asset. Counting inventory manually occupies both a person and a truck, so every counting hour is an hour the equipment is not fulfilling orders. Corvus Trident works the other direction, mounting to equipment that is already moving and capturing pallet data as operators do their normal work. It supports electric and propane forklifts, reach trucks, and order pickers.
Putaway
Putaway is the step where received goods are moved from the dock to their storage location. It is a common point of entry for inventory error, because it is the moment a pallet acquires a location in the system and everything downstream relies on that location being recorded correctly.
A putaway recorded to the wrong slot creates a discrepancy that nothing catches until someone goes looking for the pallet. Corvus Trident records the move as it happens, capturing which pallet went where from the vehicle doing the work. Corvus One catches what slipped through by covering the same locations on its next pass.
Replenishment
Replenishment is the movement of stock from reserve storage into picking locations so that pick faces stay full. It runs continuously in a busy facility and it depends entirely on the accuracy of the reserve record.
A replenishment task sent to a reserve location that is empty, or that holds something else, stalls at the rack while the pick face runs dry and somebody works out what happened. This is why reserve storage accuracy shows up in fulfillment metrics rather than inventory metrics. Several Corvus One deployments validate reserve storage specifically, which is the least visited and least frequently counted part of most buildings.
Slotting
Slotting is the decision about where each SKU lives in a facility. Fast movers near the dock and at accessible heights, slow movers further out and higher up, with the layout tuned to cut travel time on the most frequent picks.
Slotting decays. Demand shifts, seasons turn, assortments change, and a layout that was right last year quietly costs a few seconds on every pick. Reslotting is straightforward as a decision and awkward as a project, because it needs current occupancy data to plan against. Flight data showing which slots are actually full, partially full, or empty turns reslotting into an analysis rather than a walk through the building with a clipboard.
WMS (warehouse management system)
A WMS, or warehouse management system, is the software of record for what a facility holds and where. It directs receiving, putaway, replenishment, picking, and shipping, and it holds the number every other system trusts.
A WMS knows what it was told. It records scan events, so between two scans it is working from an assumption, and an assumption that goes wrong stays wrong until somebody counts. This is the gap autonomous inventory systems close. Corvus One and Corvus Trident can feed verified data back into any WMS, through a published API or standard integrations, so the system of record gets corrected by observation instead of waiting for the next manual count.