The Fleet That Doesn’t Stay: Separating the Site Layer from Vehicle Configuration

By: Luiza Bartels, Project Coordinator, U&M Mineração e Construção S.A.

What changes when the fleet leaves

Most of the literature on autonomous haulage deployment describes a single event. The mine owner specifies, buys, commissions and operates, and the cost of commissioning is amortized over years in the same pit. Under that premise, it makes sense for commissioning to be tailored to the place: the survey, the route graph, the geofences, the traffic rules, the network coverage map and the sensor calibration are built once, for a site that will not change.

U&M lives the opposite case. We are an earthmoving contractor serving the mining industry. The trucks are ours. They arrive at the start of a contract and leave at the end of it, for another mine, with another pit geometry, another mine plan, another communications infrastructure and other traffic rules. The same truck will be commissioned several times over its working life.

Since 2017 we have been developing a retrofit kit to convert off-highway trucks from our own fleet to autonomous operation, whether they left the factory prepared for autonomy or not. Our model is to deliver the technology as OPEX rather than CAPEX: clients buy no kit, no truck and no autonomy team, and they carry no obsolescence risk. They contract material moved, at a consistent standard of safety and operation.

This separates two events that the literature treats as one. Converting the truck happens once in the life of the asset. Commissioning the site happens at every contract. When the two are mixed, and they are in every deployment where the fleet stays, arriving at a new mine means repeating part of the conversion work. When they are separated, arriving at a new mine means checking the technology, running that site’s safety tests and, once approved, operating. No deployment article was found that makes such distinction, maybe because it was not necessary.

The premise of the project

U&M’s autonomous truck project was born with a premise that decided everything that came after: make no modifications to the client’s access roads and pads. If the mine already exists, we operate in it as it is.

That decision has a direct consequence. Since we do not control the geometry we will operate in, we must operate consistently across diverse scenarios. Our development effort therefore concentrated on validating complete cycles in difficult environments, rather than on validating cycles in an environment prepared for the system.

It has a second consequence, less obvious, that only shows up at mobilization. If every site arrives with the geometry it has, the network coverage it has and the traffic rules it has, none of that can be embedded in the vehicle configuration. Everything that describes the place must be replaceable without touching what describes the truck.

What belongs to the truck and what belongs to the site

The separation we maintain has two layers.

The vehicle configuration is everything that belongs to the truck and travels with it: dynamic envelope, actuation parameters for throttle, brake and steering, position and orientation of each sensor relative to the chassis, fault handling, calibration history. None of it changes when the truck changes mines, and none of it needs to be redone. In our experience, converting a 100t truck takes one eight-hour shift, followed by a month of calibration and safety testing to full commissioning. That effort is made once per truck, and the result travels with it.

The site layer is everything that describes the place: the survey and its date, the route graph, the geofences, the traffic rules, the loading and dumping points, and the communications assumptions, meaning the measured network and radio coverage with its shadow zones. That layer changes entirely with every contract.

The difficulty is not in stating the separation. It is in keeping it, because site decisions disguise themselves as vehicle decisions easily. One example: how the truck behaves when it loses communication with dispatch. How long it waits before slowing down or stopping depends on the quality of that network. If that time is written into the vehicle software, it was sized for one place’s coverage and will be wrong at the next. The same applies to a speed limit tuned to a specific ramp, or a calibration performed against a fixed reference in the mine. Each of those decisions looks reasonable while the truck stays in the same place, and each becomes rework at the next mobilization.

The pad

The most concrete case of this separation is the pad.

A large off-highway truck, to turn around in a continuous turn, needs an area sized by its turning diameter. When autonomy requires a continuous turn, the loading or dumping pad comes to be sized by the truck. If the client’s pad is smaller than that, it must be enlarged, and that civil works cost rarely appears booked as a cost of autonomy.

Since our premise was not to alter the client’s pad, we built the maneuver planner to work with reversal, combining forward and reverse segments. With it, we have maneuvered the truck, empty and loaded, within a usable area that is only slightly larger than the truck’s own turning circle clearance diameter.

What changes with reversal is that the pad stops being sized by the vehicle’s ability to turn and comes to be sized by the planner’s ability to sequence maneuvers. The pad dimension leaves the vehicle configuration and enters the site layer: it is a fact about the place that the planner receives, not a requirement of the truck that the place must meet.

Surveying the site before mobilizing

If the site layer is separated, it can be built before the truck arrives.

During the bidding phase for a service contract, the physical aspects of the mine can already be surveyed: preliminary mapping of access roads and pads, measurement of network coverage, identification of the traffic rules in force. That survey is the same one that feeds the site layer, and it can be done before any truck is mobilized.

The goal is for the mobilization of an autonomous fleet to take no longer than the mobilization of a fleet of manned equipment typically would. That is the criterion we propose for knowing whether the separation is working. If autonomous mobilization takes longer than manned, something that belonged to the site got stuck in the vehicle, or something that could have been surveyed at the bidding stage was left for the commissioning stage.

This also changes what commissioning means. Commissioning stops being the building of the site layer in the field and becomes the verification, in the field, of a site layer that arrived ready, against an acceptance procedure that is the same everywhere. That repetition is what makes the experience the team accumulates at one site count at the next, and what reduces configuration and maintenance time at every new mobilization.

The regulation already asks for the site layer

In Brazil, much of what we call the site layer already exists on paper, by legal requirement. NR-22, the Brazilian mining safety regulation, requires every mine to maintain a traffic plan with right-of-way rules, minimum distances between equipment and permitted speeds according to road conditions and, in surface mines, a plan drawing of the circulation routes and their widths, kept updated to operational need.

That plan holds most of the description of the place that the autonomous system needs. It is already mandatory, already specific to each mine and already kept current. What is missing is for it to be machine-readable. The separation between site layer and vehicle configuration is what allows a document the mine already maintains to become the system’s input, without altering either the document or the truck.

People are not commissioned again either

There is a third thing that travels with the fleet, beyond the converted truck and the site layer surveyed at bidding: the company’s knowledge of how to operate and maintain autonomous trucks.

That knowledge is already spreading through U&M beyond the development team. The operations and maintenance teams know the autonomous fleet, and the most visible effect is a substantial reduction in maintenance times. An autonomous truck that is down no longer waits for someone who understands it.

For mobilization, the consequence is direct. After commissioning at one site, there is no period of commissioning people at the next. The team that arrives already knows how to operate and maintain the system, and what must be redone is limited to what is genuinely about that place.

What is redone, and must not stop being redone, is the safety demonstration. Each mine’s traffic plan is approved for that mine, and the safety tests are run in that geometry and with that network coverage. That is what site commissioning verifies, and that is all it verifies.

What remains to be written

The separation between what belongs to the truck and what belongs to the place is a structural requirement for anyone operating their own fleet in someone else’s mine. For the mine owner, it is a convenience. For the contractor, it is the condition for arriving at a new mine to mean checking and testing, not converting again.

We have not found a deployment article that describes that boundary, perhaps because none has needed to. We think it is worth describing what belongs to the truck, what belongs to the place, and how one receives the other. It is what we have learned operating on pads and access roads we did not design, and it is what we are prepared to put up for discussion.