Before a mining truck runs on its own, someone has to walk it through three times

A friend recommended Pronto's AHS to me, saying it doesn't care what vehicle you have and can run as a retrofit. Last weekend I went to his quarry, helped install one, and tried it out to see if it's actually any good. My day job is in cockpits, not a mining person, so what I write below is just the operating process and the things I ran into—my judgment may not be right.

What AHS does is let mining trucks drive themselves from the loading point to the dumping point, dump, and drive back on their own, with the driver moving from the cab to watching a screen indoors. Pronto's approach is a bit different from others—it doesn't use the expensive and finicky lidar sensor setup; it relies on a ring of ruggedized cameras plus AI to see the road, focusing on fast installation, cheap deployment, and being vehicle-agnostic: existing mining trucks can be retrofitted and used. For a quarry unwilling to replace an entire fleet of trucks for autonomy, this approach is pretty practical.

Day one was installing the eyes. Each truck needs a 360-degree camera array, covering front, rear, and sides—essentially digitizing the whole vehicle's surroundings. Park the truck on flat hard ground, shut off the engine, disconnect the negative battery terminal. Mount the cameras per the bracket positions, orient them per the manual—don't go by feel. The first time I mounted one two or three degrees off, and the stitched image later was misaligned. Connect the main unit and wiring harness, route the cables away from the exhaust pipe and spots where they'll get knocked, then power on, open the console, go to Vehicle Management, and see if this truck shows up on its own—status should be "Online / Pending Calibration." The expected result is the truck being recognized by the console. The first pitfall is right here: mud spots on the lens. The system won't report an error—it just won't pass calibration no matter what. Wipe it clean and redo it. Industrial sites are dusty; this is something you'll do every day from now on.

Day three is taking it on a walk-through—this step is called teaching, which in plain terms means you drive it manually once so it remembers the route. First create a new site in the console under Site Management, give it a name, then drive a truck around the site, passing through the main road, loading point, dumping point, and fueling spot—it builds the map as it drives. Back on the map, click Mark Loading Point, Mark Dumping Point, name them one by one, then go to Route Teaching, manually drive from the loading point to the dumping point, stop steadily, and click Save Route. Repeat two or three times for stable localization. Second pitfall here: the dumping point should be marked at the position where it will reverse to later, not parked head-in for convenience. The first time I parked head-in, and it ended up half a truck body off every time it reversed—had to re-teach. Temporary stockpiles on the site should either be cleared or marked as obstacles on the map, otherwise it will dutifully drive right into them before emergency stopping.

A week later, let it run on its own. Switch to auto, in Tasks select the truck, select the route, fill in the number of trips, click dispatch, and then you just watch it run. The three numbers I watch are: number of takeovers, lap time, and the number of times it inexplicably stops. Takeovers are the most honest—every time a person goes up to take over, that's one. The third number is also interesting: often when it stops, it's because the connection between the console and the truck dropped. When the connection drops, the truck stops in place and waits—it won't run wild. That's a safety design, but the first time you encounter it you'll think it's broken.

A few more general pitfalls. Cameras struggle with backlight—best to check the image when the morning sun is low. If the site changes a lot, e.g., a new road is opened, the map needs rebuilding, otherwise localization will drift. Don't expect to go unmanned the same day you finish installing—these fiddly tasks can't be skipped, not even for a day. Pronto has already deployed over a hundred trucks at Heidelberg, running over two million tons, and will enter two new sites in 2026. That shows it can run at scale, but that's someone else's site—your site will have every pitfall to step in.

Once a single truck runs smoothly, you can try multi-truck scheduling, see whether two trucks will wait for each other on the road. Beyond that is mixed traffic—letting autonomous trucks and human-driven trucks run in the same site. That step is much harder than single-truck autonomy. Whether the site is willing to change practices they've used for over a decade is a harder hurdle than the technology.