Logistics

Fusing Forklift Cameras with UWB Tags for Pallet-Level Accuracy

An event-driven recipe: carriage cameras identify pallets, UWB locates the truck, and lift events bind pallet IDs to rack slots with no pallet tags.

Hayat Amin, President of IP, Position Imaging Hayat AminPresident of IP, Position Imaging 4 min read
The short answer

Put the camera and the UWB tag on the forklift, not the pallet. The camera reads the pallet label at pickup, UWB locates the truck to 10 to 30 cm, and a load sensor's lift and drop events bind pallet ID to rack slot. Every put-away then writes its own inventory record, so pallet-level accuracy comes from a few instrumented trucks instead of thousands of tagged pallets.

Key takeaways

  • Tag the truck, not the pallet: one UWB tag per forklift replaces thousands of disposable pallet tags.
  • Fusion is event-driven: lift and drop events from a load sensor are the join key between camera-read pallet IDs and UWB positions.
  • Project position to the fork tips, not the tag: 5 degrees of heading error at a 2 meter lever arm costs 17 cm.
  • A mast height encoder, not the radio, resolves which rack level the pallet landed on.
  • Keep total lateral error under about 45 cm and put-aways snap to the correct slot in a 0.9 to 1.4 meter slot pitch.
  • Manual pallet-jack moves are the biggest coverage gap, so design exception workflows on day one.

Why can't cameras or UWB alone track pallets?

Start with the two failure modes. A UWB tag on every pallet looks clean on a whiteboard until you price it. Tags run roughly 10 to 25 dollars each with a battery to manage, and pallets churn through a busy building by the hundreds of thousands per year. Vertical accuracy inside steel racking is the other problem: UWB ranging handles the horizontal plane well, but a tag at level 4 and a tag at level 2 can return near-identical ranges to the same ceiling anchors, and steel beams multiply multipath.

Cameras have the opposite gap. A fixed camera network cannot see into a rack bay once product fills it. A forklift camera can decode a GS1-128 label at half a meter, but on its own it has no idea which of 4,000 slots it is facing, and visual odometry drifts because warehouse aisles look alike.

The camera knows which pallet. The radio knows where the truck is. Each sensor answers half the question.

What goes on the truck: the minimum fusion stack

Four components per vehicle, all commodity hardware:

  • One UWB tag on the overhead guard, running 802.15.4z HRP ranging at 5 to 10 Hz against ceiling anchors spaced 20 to 30 meters along aisles. Expect 10 to 30 cm horizontal accuracy with line of sight to three or more anchors.
  • A carriage-mounted camera with a global shutter, aimed at the fork face, decoding the pallet label at the 0.5 to 2 meter engagement distance. Global shutter matters because the mast vibrates.
  • A mast height encoder or wire draw sensor reporting fork height to a few centimeters. This is what resolves the rack level the radio cannot.
  • A load or lift sensor on the carriage, so the system knows the exact moment the forks take or release weight.

An edge computer timestamps all four streams. Keep clocks within about 50 ms, because a truck moving at 4 m/s smears 20 cm of position error into every event for each 50 ms of skew. Four sensors, one truck, zero pallet tags.

How does a lift event bind a pallet to a slot?

The fusion is event-driven, not continuous. You do not track pallets; you track custody changes.

At pickup, the load sensor fires, the camera decodes the license plate on the pallet label, and the system writes one record: pallet P is on truck F at time t. From that moment, the pallet's position is simply the truck's position.

At put-away, the load sensor fires again. Now compute where the fork tips are, not where the tag is: take the UWB position, apply the truck's heading, and project forward along the fork offset, typically 1.5 to 2.5 meters from the tag. Add mast height from the encoder. Snap the result to the nearest slot in the rack map. Slot pitch runs about 0.9 to 1.4 meters laterally and 1.8 to 2.4 meters vertically, so a total lateral error budget under 45 cm assigns the correct slot nearly every time. Heading is the silent killer in that budget: 5 degrees of error at a 2 meter lever arm costs 17 cm before the radio contributes anything.

The lift event is the join key.

Which failure modes break pallet-level accuracy?

These are predictable, so budget mitigations before the pilot:

  • Manual moves. Hand pallet jacks and uninstrumented reach trucks create ghost records: the system still believes the pallet sits where the forklift left it. Either instrument every vehicle that touches a pallet or route uncovered moves through a scan-confirm exception workflow.
  • Unreadable labels. Shrink wrap glare and torn labels defeat a share of decodes. Carry the last confirmed pallet ID through the trip, and flag any drop with a low-confidence read for a one-tap worker confirmation rather than writing a bad record.
  • Double-deep and drive-in racking. The fork-offset projection must include rack depth, or the front and rear positions of a double-deep bay collapse into one slot.
  • End-of-aisle steel. UWB error grows where racking, dock doors, and mezzanines crowd the geometry. Place anchors so a truck sees three with line of sight at aisle entrances, where turns concentrate heading error.

Plan for the pallet jack first.

Should you build this fusion stack or license it?

The sensors are commodity parts you can buy this week. The hard part is the method: associating an optically identified object with an RF-located vehicle through a custody event, projecting through a lever arm, and keeping the record store consistent when reads fail. That method space is heavily patented, and infringement risk surfaces after you ship, when fixing it is most expensive.

Position Imaging holds hundreds of granted patents across RF ranging, computer vision, and machine-learning tracking, a portfolio cited by Apple, Bosch, and other major filers. Granted patents relevant to this architecture include US 11,774,249, US 12,079,006, US 12,066,561, and US 12,000,947. Licensing buys two things at once: a documented head start on methods that already work in production, and freedom to operate so legal review does not stall your rollout. Teams that license proven positioning IP typically get an instrumented-truck pilot running in months, against the 18 to 24 a from-scratch fusion effort tends to consume.

Buy the method, build the product.

Patents referenced
US 11,774,249US 12,079,006US 12,066,561US 12,000,947

Frequently asked questions

Do I need a UWB tag on every pallet for pallet-level inventory?

No. Tag the forklift, not the pallet. The carriage camera identifies the pallet at pickup, and the truck's UWB position at the drop event records where it landed. One tag per vehicle covers every pallet that vehicle touches, which is why the economics work at hundreds of thousands of pallet moves per year.

How accurate is UWB on a moving forklift?

With 802.15.4z ranging at 5 to 10 Hz and line of sight to three or more anchors, expect 10 to 30 cm horizontally. The number that matters more is the error at the fork tips after projecting through a 1.5 to 2.5 meter offset, where heading error adds roughly 3.5 cm per degree at a 2 meter lever arm.

How does the system know which rack level a pallet went to?

Not from the radio. UWB resolves vertical position poorly inside steel racking, so the rack level comes from a mast height encoder or wire draw sensor read at the moment the load sensor fires. With rack levels spaced 1.8 to 2.4 meters apart, a few centimeters of encoder accuracy is more than enough.

What happens when the camera cannot read a pallet label?

The system carries the last confirmed pallet ID from the pickup event through the trip, since the pallet cannot change while it is on the forks. Drops with low-confidence or missing reads get flagged for a quick worker confirmation instead of silently writing a bad slot record.

How many UWB anchors does a warehouse need for forklift tracking?

Plan anchors every 20 to 30 meters along traveled aisles, placed so a truck keeps line of sight to at least three at any point, with extra density at aisle entrances and dock areas where turns and steel degrade geometry. A 20,000 square meter building typically lands in the range of a few dozen anchors.

Talk to the IP team

Send us your truck and rack specs, and we will map your fusion architecture to the specific granted patents it touches.

Tell us the product. We map the exact scope, what a license covers, and how fast you can ship, all in a 20-minute call.

Book a 20-minute call