There was a stretch around 2022-2023 where every mining conference panel sounded the same: autonomous haulage was the future, productivity gains were enormous, and any operator not moving was being left behind. The marketing was ahead of the data.

By mid-2026 we have something closer to honest numbers. Several Pilbara operators have been running fully autonomous truck fleets for long enough that the maintenance, uptime and incident data is statistically meaningful. Let’s look at what it actually says.

Uptime is good, but not the headline figure

Across the major Pilbara iron ore operations, autonomous truck availability is running between 89 and 93 percent on a rolling twelve-month basis. That’s a few percentage points above comparable manned fleets, which sounds like a clear win. The catch is what counts as available.

A truck that’s parked because the central control system has lost confidence in its lidar reading is “available” in some operator definitions and “down” in others. The reporting standards still vary, and any benchmark comparison needs to account for that. The Australian Centre for Robotics has done useful work trying to standardise these definitions, but adoption is uneven.

Where the data is consistent, the genuine uplift over manned fleets sits closer to two to four percentage points of utilisation, not the eight-plus that early business cases assumed. That’s still meaningful at scale - on a 60-truck fleet running 24/7 it’s hundreds of millions of tonne-kilometres of additional capacity per year. It’s just not the step-change number some boards were sold.

Tyre wear and the routing trade-off

One genuinely surprising finding from operator reports has been the tyre cost story. Autonomous fleets follow optimal speed and line profiles consistently, which sounds like it should reduce tyre wear. In practice it does the opposite for some pit configurations.

Because every truck takes precisely the same line through corners, the wear pattern on the haul road concentrates rather than distributing. This accelerates road maintenance cycles and increases tyre cuts from the harder, polished surface. Several operators have started programming deliberate small variations in the autonomous route to spread the load. Others have invested in road maintenance fleets to compensate.

The net cost picture is still favourable - autonomous fleets save more on labour, fuel optimisation and incident-related downtime than they spend on tyres and roads. But the original business cases that assumed tyre wear would improve have had to be rebuilt.

Mixed-mode operations are still the hard problem

The cleanest autonomous numbers come from greenfield pits designed around the technology - controlled access, no light-vehicle traffic, no manned trucks sharing the route. That’s not where most of the industry’s ore comes from.

Brownfield deployments mixing autonomous and manned equipment continue to underperform pure-mode operations on every metric. The handover protocols, the geofencing of areas where humans need to work, the management of contractor light vehicles - these are operational problems that haven’t been solved with software, only with discipline. Operators that have invested heavily in change management and operating-procedure rewrites are seeing the gains. Those that bolted autonomous trucks onto an existing operation and hoped culture would follow are not.

Caterpillar’s recent technical paper on mixed-mode safety contains some sobering data on near-miss frequencies in mixed environments. The numbers are not catastrophic, but they’re high enough that any operator considering retrofit autonomy should plan for a multi-year cultural transition, not a six-month deployment.

Where the AI overlay is actually paying off

The most interesting recent development isn’t autonomy itself - it’s the AI dispatching layer that sits above autonomous and manned fleets together. Several operators are running optimisation systems that reroute trucks in real time based on shovel queue dynamics, road conditions, and downstream stockpile demand.

The reported gains here look meaningful: 4-7 percent improvement in tonnes-per-shift on the operations that have rolled it out properly. That’s bigger than the gain from autonomy itself in many cases. The Australian operators leading this work have been quietly partnering with local AI specialists - groups like Team400 and similar consultancies are showing up in the technical credits of more papers than they were two years ago, particularly on the optimisation and operations-research side.

Labour, the unspoken story

The labour piece deserves a paragraph that operator press releases generally skip. Autonomous deployment has not eliminated mining jobs at most operations - it has shifted them. Truck drivers have been replaced by remote operators, system technicians, road maintenance crews, and a layer of analytics and operations engineers that didn’t exist a decade ago.

The headcount on a typical autonomous Pilbara operation is roughly 70-80 percent of the equivalent manned operation, not 30-40 percent as some early forecasts suggested. The skill mix has shifted significantly, and the regional employment impact has been smaller than either supporters or critics predicted.

The 2026-2028 trajectory

Three trends look durable. First, the standalone “autonomy as productivity miracle” story has matured into something more nuanced and more honest. Second, the AI optimisation layer above the equipment is where the next wave of gains is being captured. Third, the gap between operators who treat autonomy as a cultural and operational programme versus those who treat it as an equipment purchase keeps widening.

If you’re benchmarking your own operation against autonomous deployments, look past the availability headline and into the road maintenance, tyre, and incident data. That’s where the truth sits.