Simulated coil yard · 2 cranes · representative bay geometry

← goatai.io/steel

YardVision

As the hook descends, where is it relative to the coil it's about to pick up?

AI-native yard intelligence for slab and coil yards. One OptIMU pod on each crane — nothing on the runway, nothing in the hook — resolving the axis legacy sensors never did: the moment of pick-up and lay-down.

Steel Yards Today

Three problems, three add-ons

For decades, crane and yard tracking has meant bolting on more sensors. Heat, dust, vibration and misalignment keep maintenance busy — and in daily operation the data has never been reliable enough to trust.

Laser & radar

Distance meters and radar for long and cross travel, with reflectors along the runway. Real yard geometry drifts the beam; heat, dust and vibration do the rest.

Load cells

Hook load sensing just to know when a coil was picked up or laid down. When one fails, recovery waits for a shutdown window — and the yard runs blind on its most critical event.

The Z-axis

Which tier, which level? Long and cross travel got instrumented. Height — the moment of pick-up and lay-down — stayed with the operator's eye.

Many sensors. Many failure points. And the most important event in the yard — pick-up and lay-down — still unseen.

The Pod

Meet OptIMU — vision and inertial sensing, fused in one pod at the cabin

Instead of a sensor per axis, one pod per crane that understands the space around it. We are targeting 5–10 cm hook localisation, and the field question is how consistently it holds across real pickups.

Replaces laser & radar

Spatial perception

OptIMU understands the space around the crane: long travel, cross travel and hoist height, including which tier. No lasers, no radar, no reflectors on the runway.

Replaces load cells

Pick-up and lay-down, seen

Every grip and release is recognised and binned to lane, row and tier. Nothing in the hook to fail, and nothing to wait a shutdown for.

Trust by design

It knows when it is not sure

Confidence is tracked on every event. Doubtful ones go to the operator instead of silently into the record.

Trust by design

And recovers on its own

Column ID plates and structural fixes re-anchor position automatically. Back to centimetres without recalibration or a site visit.

5–10 cmhook localisation target relative to the coil or slab being picked — a target to be proven in the field, not a demonstrated result.

A Different Approach

Legacy sensors vs AI-native

Legacy sensor stackYardVision
HardwareLasers, radar, runway reflectors, hook load cellsOne OptIMU pod per crane
AxesLong and cross travel; height is the weak spotLong travel, cross travel, hoist height — plus pick and lay-down events
On the runwayTargets and reflectors to align and maintainNothing to install or align
IntelligenceRaw distances and weightsPosition, events and confidence
When it's unsureWrong values look the same as right onesConfidence on every event; doubtful ones go to the operator
OutputPositionsA live coil twin, dispatch sequencing and walkway lockout

Yard Simulator

Run the yard before you instrument it

Physics-based slab and coil yards with EOT cranes, live traffic and a YardVision camera kit on each crane. Switch between yards, degrade detection, add IMU drift — and watch the live twin hold or ask the operator.

Eye-horizontal coils in two tiers, transfer cars in, trucks out in dispatch order, walkway lockout under a suspended coil. Best on a desktop screen. Bay geometry is representative of a 45 t × 34 m span DG EOT bay; column, truss and crab detail are schematic.

Simulated coil yard · 12 h of yard operation · 2 cranes · 2,215 coil slots

0

silent mis-bins

768 pick and lay-down events. None went silently into the record.

91%

binned automatically

The other 9% were sent to the operator to confirm.

−40%

forced reshuffles

12 vs 19 in the same 12 h — dispatch-aware vs random placement.

Simulation results, not field measurements. They show the method — confidence-gated binning and dispatch-aware placement — not performance in a live yard.

Where It Comes From

Positioning accuracy depends on the yard as much as the sensor

YardVision comes out of more than two decades of putting lasers, RF and radar on steel-plant cranes — and learning what each of them couldn't do.

1998

Laser, and the 2 m reflector

A laser distance meter put into service for overhead-crane positioning. It worked — then showed that the yard's real geometry wasn't what the drawings assumed. Holding the beam on target down the aisle needed a reflector about 2 m × 2 m.

2008

RF and radar ranging

RF ranging promised a way through the dust and an end to large optics. Radar later improved ranging in a coil yard; at another site a radar yard system struggled with vibration and alignment and was taken out.

Since

The axis that was never resolved

Along the aisle and across the bay could be tracked. The vertical — pick-up and lay-down — stayed with the operator's eye or a load cell. Years of shop-floor vision work made it time to go back to the yard.

The lasting lesson: real yard geometry and how a sensor is mounted matter as much as the sensor itself. YardVision starts where the old systems fell short — the vertical.

Safety Architecture

Safety is physical. Vision does the yard work.

Remote yard operation has historically stalled on asking the perception sensor to also prove no human is present. YardVision separates the two: a sealed cell with hardwired gate interlocks owns human safety; vision is freed for machine geometry, load position and clearances.

Layer 1 · Interlock

Boring on purpose, highest integrity

Gate-state — not person-detection — is the interlock. Gate open → crane locked out, whether or not anything is seen. Fail-safe by construction and independent of the perception stack: if vision has a bad day, the interlock still fires.

Layer 2 · Vision

Unburdened

OptIMU runs crane geometry, hook localisation, load position and clearances. Vision performance is stated for the machine-and-load task — explicitly not for human safety. Walkway lockout under a suspended load is an additive layer, never the thing life depends on.

01

Yard monitoring

Scope

Crane position, hook localisation, pick-up and lay-down events, and a live coil and slab twin. Sense, log, alert — human in command.

02

Operator assist

Scope

Placement guidance, clearance warnings and walkway lockout to the cab. The system advises; it does not act.

03

Sealed cell + gate interlock

Built toward

Gate-open → all-crane-stop, hardwired, fail-safe and perception-independent. This rung gates everything above it.

04

Remote operation

Built toward

Cabin-out control of a verified-empty cell. Vision is the operator's eyes for machine and load; the interlock owns human safety.

05

Bounded autonomy

Built toward

Routine, low-consequence moves inside the interlocked cell, supervised and overridable.

Prove It In Your Yard

Co-create YardVision's first field proof

We're inviting steel plants across India to define and run the first field proof of concept with us — tested together, in your yard, against what your team considers useful.

One crane

An OptIMU pod on one existing crane. Nothing on the runway, nothing in the hook.

One bay

Shadow mode alongside current operations — no change to how the yard runs.

One recurring task

A pickup your team does every shift. Your operators and maintenance team define what a useful result looks like.

How does your team confirm pick-up and lay-down today — the operator's eye, a load cell, or something else?