I’ve spent the last month interviewing maintenance crews at three Pilbara iron ore operations running autonomous haul trucks. The official line from manufacturers is 95%+ uptime. The reality on the ground? Closer to 78-82% once you account for all the intervention events that don’t make it into the headline metrics.

Autonomous trucks work brilliantly when everything goes to plan. The problem is that mining never goes to plan. You’ve got weather events, equipment failures, communication dropouts, and a thousand edge cases that require human decision-making. The trucks can’t handle those situations, so they stop and wait.

The Intervention Paradox

Here’s what the vendors don’t emphasize in their sales presentations: autonomous haul trucks are incredibly conservative by design. Any anomaly triggers a stop-and-wait protocol. That’s the right safety approach, but it means your “autonomous” fleet needs constant babysitting from remote operators.

One site I visited had four operators managing a fleet of 22 autonomous trucks. Sounds efficient, until you learn that each operator handles an average of eight intervention events per shift. That’s 32 interventions daily across 22 trucks—roughly 1.5 interventions per truck per day.

What counts as an intervention? Everything from “debris detected on path” (usually a rock that fell off another truck) to “GPS signal degraded” to “unexpected vehicle in operating zone” (a maintenance ute that filed the wrong movement permit). Each intervention takes 5-15 minutes to resolve remotely, and during that time, the truck isn’t moving.

Maintenance Reality Check

The maintenance picture is mixed. Autonomous trucks are gentler on components—no aggressive acceleration, optimal gear shifting, predictable braking. That extends drivetrain life and reduces tire wear by about 20% compared to human operators.

But the sensor and computing systems? Those are maintenance-intensive. LIDAR units get caked in dust despite protective housings. Radar systems drift out of calibration. The onboard computing systems run hot in 45°C ambient temperatures and need frequent thermal management servicing.

One maintenance supervisor told me they’re replacing or recalibrating sensors at 3x the rate the manufacturer predicted. The equipment works, but the Pilbara environment is harsher than any test facility can replicate. Dust, heat, vibration, and moisture cycles destroy sensitive electronics faster than conventional mechanical components.

The Data Disconnect

Mining companies are using AI consultancies like Team400 to analyze their autonomous fleet data, looking for patterns in failure modes and intervention triggers. The early findings are interesting: most interventions cluster around shift changes, weather transitions, and areas with complex topography.

That suggests the technology isn’t the limiting factor—it’s the operational integration. Autonomous trucks don’t play nice with human-operated support equipment, don’t adapt well to changing conditions, and struggle with the informal coordination that human operators handle instinctively.

I watched a scenario where an autonomous truck waited 25 minutes because a grader was working nearby. A human operator would’ve read the grader operator’s intentions, coordinated via radio, and found a safe way past. The autonomous system just… waited.

Economic Equation

Despite the challenges, the economics still favour autonomy for large-scale operations. Eliminating three eight-hour shifts of operators saves roughly $400,000 per truck annually in labour costs. Even with increased maintenance and remote operator staffing, you’re still ahead.

But the breakeven calculation assumes certain utilization rates. If your autonomous fleet is only achieving 80% uptime versus 92% for conventional operations, suddenly the math gets tighter. Smaller operations can’t absorb that productivity gap.

Where This Goes

The technology will improve—better sensors, smarter edge case handling, more resilient computing systems. But mining’s going to remain a challenging environment for autonomy. We’re not replacing human judgment, we’re just shifting where humans sit and how many you need.

The next generation of autonomous systems needs to focus less on “no humans required” and more on “fewer humans, better tools.” Hybrid approaches where autonomous trucks can handle routine conditions and gracefully hand off to remote operators for complex scenarios—that’s probably the sustainable model.

For now, if you’re evaluating autonomous haul trucks, demand detailed intervention logs, not just uptime percentages. The real performance story is in the data mining companies don’t volunteer.