What mining can borrow from the industries that already have a positioning layer

By: Aaron Nathan, CEO and Founder, Point One Navigation

A haul truck from one OEM and a drill rig from another can both report centimeter accuracy on the same bench, in the same shift, and still disagree about where the bench actually is by half a meter or more. Nobody notices until a blast hole comes in off pattern, a load gets reconciled to the wrong stockpile, or two vehicles running two different autonomy stacks both believe they own the same fifty meters of haul road. The receivers were never the problem. Centimeter accuracy meant something different inside each vendor's box, and nothing forced the boxes to agree.

I've watched this exact failure mode play out, with different props, in four other industries, years before mining inherited the same problem. In every case, the fix wasn't a better chip or a more clever algorithm. It was an industry-wide decision to stop letting each vendor define positioning on its own terms: pull it out of the vehicle or the instrument, and turn it into a layer, with a standard interface anyone can build against and an accuracy and integrity budget that's specified once and honored by whatever plugs in behind it, no matter who built the box.

Automotive made that move. Agriculture made it twice, once for the machines and once for the corrections feeding them. Surveying made it before either of them existed as automated disciplines, mostly because it had no choice. Logistics made it more recently, for close to the reason mining will eventually need to: mixed-vendor robot fleets don't run under one control system until somebody defines the seams between them. Mining hasn't made that move yet. It's re-solving the identical problem inside every vendor silo it buys, which is an unusual place to be, because the answer already exists, running in production, in industries with harder liability exposure and just as much reason to distrust a competitor's spec sheet.

The vantage point behind this piece is a narrow one. The positioning stack under a delivery van, a spray robot, a survey rover, and a passenger vehicle that has to answer to a regulator can be, and in our case is, the same interface held to the same budget. Amazon runs it across 100 percent of its Last Mile delivery fleet. Agra-GPS and GUSS build it into agricultural guidance and spray robots. Carlson and Civ Robotics put it under survey equipment. Zoox puts it in vehicles that answer to a regulator or a safety case, not just a spec sheet. Nine years of moving the same interface between those industries is the actual credential behind what follows, not any single deployment on its own.

Why mining keeps re-solving positioning

I wrote here a few months ago about fleet-level position management, and about how ISO 23725's flexible boundary model puts positioning at the intersection of the machine, the Fleet Management System (FMS), and the site, without assigning it to any one of them. That ambiguity is deliberate, and it's the right call for how the standard treats trajectory generation and traffic rules, which genuinely need to flex across different vendor architectures. But positioning inherited the same ambiguity by default, and nothing in the standards, or in most sites' procurement process, has closed that gap the way the market eventually closed it everywhere else.

Every other industry on this list had something mining hasn't had yet: a forcing function that made positioning too expensive, too dangerous, or too commercially awkward to leave bundled inside a single vendor's box. Automotive got functional safety regulation. Agriculture got a farmer who owns three brands of equipment and wants one screen in the cab. Surveying got professional liability. Logistics got a warehouse operator who refuses to run four separate fleet controllers for four brands of robot. Mining is mid-transition: most Autonomous Haulage Systems (AHS) deployed today are still single-OEM, closed stacks, which sidesteps the interface problem the same way a closed automotive platform would, by not exposing it. Some vendors are beginning to shift this dynamic. Pronto, for instance, has focused on interoperable off-road autonomy intended to work across mixed-vendor sites. That works until a site wants a second AHS vendor, a different drill guidance provider, or a positioning layer it can actually audit, at which point the bundled approach turns into the same silo problem the other four industries already paid to get out of.

Part 1: Automotive, a number for how wrong the system is allowed to be

Automotive didn't solve positioning by making it more accurate. It solved it by making the acceptable failure rate an actual number instead of an adjective. ISO 26262, the functional safety standard for road vehicle electrical and electronic systems, assigns every safety-relevant function an Automotive Safety Integrity Level (ASIL), from A to D, based on a formal hazard analysis of severity, exposure, and controllability. GNSS positioning feeding a lane-keeping, automatic emergency braking, or automated driving function typically has to clear ASIL B or higher, and the probabilistic failure-rate most teams design ASIL D systems against is a ceiling of 10 failures in time, or 10 failures per billion hours of operation. That number doesn't live on a marketing slide. It lives in a hazard analysis, a verification plan, and an audit trail that a supplier can hand to an OEM regardless of whose GNSS receiver or correction service sits behind it.

What transfers to mining isn't the specific ASIL framework, which is a road vehicle standard and doesn't map cleanly onto a haul truck's operating envelope. What transfers is the habit: an accuracy number by itself says almost nothing about how a system behaves in the tail. A positioning excursion near a loaded haul truck, a blast exclusion zone, or a light vehicle sharing a ramp is closer to an ASIL-B hazard than to a spec-sheet footnote, and it deserves the same treatment, a stated failure rate, tied to a hazard analysis, that an AHS vendor and a positioning provider can both be held to, independent of either one's own accuracy claim.

Part 2: Agriculture, an accuracy vocabulary that travels

Agriculture solved this from two directions at once. ISO 11783, known as ISOBUS, standardizes how a tractor, a planter, a sprayer, and a display talk to each other regardless of manufacturer, so a farmer can run a mixed-brand fleet off one terminal instead of a different proprietary box for every implement. Work on it started in 1991, it reached tractors and implements in 2001, and the Agricultural Industry Electronics Foundation certifies compliant equipment today. Separately, and just as consequentially, agriculture standardized how it talks about accuracy. Guidance vendors don't publish a single accuracy number. They publish pass-to-pass accuracy, how far the system drifts from a line it drew fifteen minutes ago, at a 95 percent confidence interval, and year-to-year accuracy, whether it can find that same line again next season. A typical RTK guidance spec reads something like plus or minus one inch pass-to-pass at a defined baseline distance, under stated sky and interference conditions, because a farmer running controlled traffic or inter-row cultivation needs the actual statistical behavior of the system, not its average.

What transfers to mining is that second habit more than the first. A site buying grading, drilling, or line-of-sight-critical positioning should be asking for the mining equivalent of pass-to-pass and year-to-year numbers, at a stated confidence level, under stated conditions, portable across whichever AHS or drill guidance vendor is running that fleet this year. "Centimeter accuracy" on a spec sheet is the mining industry's version of the number agriculture walked away from thirty years ago.

Part 3: Surveying, where most of this interface actually came from

Surveying didn't invent this pattern, but it's the field that proves how old it is. RTCM, the correction message format underneath RTK on every platform in this article, originated in 1983 with the Radio Technical Commission for Maritime Services' Special Committee 104, formed to standardize differential GPS corrections for harbor and coastal navigation, years before RTK existed for anything on land. Land surveying was among the first fields to adopt it wholesale for centimeter-level positioning, and NTRIP, the internet delivery protocol that makes those corrections reachable from any receiver on any network, was standardized specifically to extend that same interface beyond radio beacons. Every industry in this piece, mining included, runs corrections over an interface that started in maritime navigation and was pulled into centimeter-level accuracy by surveying, not by any GNSS receiver manufacturer.

Surveying also attached the sharpest accuracy and integrity budget of the four. The ALTA/NSPS Land Title Survey Standards, jointly published by the American Land Title Association and the National Society of Professional Surveyors, cap the allowable Relative Positional Precision on a boundary survey at 2 centimeters plus 50 parts per million, computed at a 95 percent confidence level from the error ellipse between adjacent corners. This is a liability threshold: exceed it, and the surveyor has to document why, in a record that underwrites a title insurance policy.

What transfers to mining is the discipline of treating the reference network itself, not just the rover, as the thing that needs a documented, portable accuracy budget, and the willingness to express that budget as a real statistical quantity instead of a headline figure. What doesn't transfer directly is the cadence: a boundary monument is expected to sit still for years. A reference point in an active pit often isn't, and I'll come back to that.

Part 4: Logistics, one fleet brain, any vendor's robot

Logistics solved a version of mining's exact interoperability problem, just indoors and a decade earlier. Warehouses running mixed fleets of Automated Guided Vehicles (AGVs) and Autonomous Mobile Robots (AMRs) from different manufacturers used to need a separate control system per brand. VDA 5050, developed by the German Association of the Automotive Industry (VDA) and the German Mechanical Engineering Industry Association (VDMA), defines a standard interface, over MQTT and JSON, between any compliant vehicle and a single master control. A robot's state message, its position, its battery, its order status, looks the same to the fleet manager whether the robot underneath it came from one vendor or five. The current version, 2.1.0, shipped in January 2025 with explicit extensions for larger, more heterogeneous fleets, and version 3.0.0 followed in March 2026 with path-sharing and zoning for freely-navigating vehicles, which tells you the standard is still being pulled forward by exactly the mixed-fleet pressure mining is starting to feel.

It's worth noting who wrote it. VDA, the German Association of the Automotive Industry, is the same industry whose functional safety discipline shows up two sections back as ISO 26262. The habit of writing down an interface and holding every vendor to it crossed from cars into warehouse robots before it crossed into mining.

What transfers to mining is closest to a template here, because ISO 23725 is already doing for mining's FMS-AHS handshake roughly what VDA 5050 does for a warehouse's master control. The gap is that VDA 5050's state message was designed with position and fleet coordination in mind from the start. Mining's interface got there for dispatch and safety messaging and left the positioning contract for individual OEMs to define, which is the reason this article exists.

Translating all of this for a mixed-fleet mine

Strip the industry-specific detail out of all four use-cases and the transferable pattern is short: define one interface for corrections delivery and position and state reporting, and use it for every device regardless of vendor. Attach an explicit accuracy and integrity budget to that interface, expressed as a percentile or a confidence interval under stated conditions, instead of an average. Decouple the correction source from the platform, so the interface looks identical whether the corrections come from a shared network or a base station on site. None of that requires picking a winner among AHS vendors, GNSS receivers, or correction providers. It requires deciding, once, what "accurate enough" and "wrong in a way we'd catch" mean for a given site, and holding every vendor to that definition instead of whatever their own box happens to report.

The specific budget a mine should borrow is a hybrid, not a straight import from any one of the four. The interface and the accuracy vocabulary should look more like agriculture and surveying: portable, statistically defined, vendor-agnostic. The integrity discipline, the actual failure rate behind that budget, should look more like automotive, because the consequences of getting it wrong sit closer to a haul truck excursion than to a crooked planting row.

Where a pit isn't a highway

None of this transfers cleanly, and pretending otherwise is how a site ends up importing assumptions that don't hold on the ground.

Remote sites break the delivery assumption, not the interface. Automotive, most of agriculture, and most logistics assume a correction source reachable over commodity cellular. Plenty of pits can't assume that. The lesson isn't "use a public network." It's that the interface, the mountpoint model, the credentialing, the accuracy budget, should look identical whether the correction source is a shared network or a single base station on a bench, so a remote site isn't a different integration from a connected one.

The reference station itself isn't permanent. A CORS monument or a farm's local base is expected to sit still for years. A base station on an active bench has a shelf life measured in blast cycles. Borrow surveying's discipline of verifying control against known points, but tie the re-verification cadence to the mine plan, not the calendar. A reference point that moved when the ground under it moved is a datum problem waiting to show up as a positioning error nobody can explain.

GNSS-available and GNSS-denied aren't two problems here. They're one interface. A pit transitions between open sky on a bench and blocked sky against a highwall or down a ramp as routine operation, not an edge case the way a tunnel is for automotive. That's closer to what an indoor logistics fleet or a survey crew running total stations already had to design for than to anything automotive treats as a default condition. The positioning interface has to carry a valid confidence number across that handoff, not just a GNSS fix state that goes meaningless the moment the sky disappears. This is why sensor fusion is so important in mining environments, a topic for another time.

The standard exists for messaging, not yet for positioning, and most mines are still single-OEM. ISO 23725 hasn't been tested by the same mixed-fleet pressure that forced VDA 5050's hand, because most AHS deployments today are still one vendor's closed stack. That won't hold for every site running a multi-decade mine plan. The sites that design the positioning interface now, while a change is cheap, won't be the ones re-architecting it under a second AHS vendor's onboarding deadline.

A framework for making positioning a layer, not a silo

1. Define the interface once, in writing, for corrections delivery and position and state reporting, and hold every device on site to it regardless of AHS, drill guidance, or grade control vendor.

2. Attach an explicit accuracy and integrity budget to that interface, expressed as a percentile or confidence interval under stated conditions, borrowing agriculture and surveying's vocabulary and automotive's rigor.

3. Decouple the correction source from the platform, so a base station on a bench and a shared network feed the same interface the same way.

4. Tie re-verification to the mine plan, not a calendar, because the ground your reference points sit on is the one thing none of the other four industries had to worry about moving.

5. Build the interface before a second AHS vendor forces it. ISO 23725 gave mining the handshake. It didn't give it a full positioning stack.

Automotive, agriculture, surveying, and logistics already paid the tuition to learn that positioning works better as infrastructure than as a feature bundled inside somebody else's box. Mining is the only one of the five still paying it a vendor at a time, but not for long.