Breaking down silos: improving coordination in autonomous mining

Mining has spent much of the past decade automating individual tasks. Driverless haul trucks now transport ore continuously across some of the world's largest open pits. Autonomous drill rigs execute blast plans with centimetre precision. Inspection drones survey pit walls, stockpiles and haul roads faster than conventional surveying methods, while AI increasingly supports predictive maintenance, dispatch and safety monitoring. 

Yet despite those advances, much of that automation remains fragmented.

A haul truck may optimise its own route without knowing that a drone has detected changes to pit geometry. A maintenance platform may predict equipment failure without sharing that information with fleet dispatch software. Each system performs its designated function efficiently but often with limited awareness of the wider operation.

Industry observers increasingly see fragmentation as the next great challenge for mining automation. GlobalData’s recent report, Autonomous Mining: Unlocking Innovation and Competitive Advantage, identifies a transition towards “fully integrated, AI-driven fleet orchestration”, combining sensor fusion, predictive maintenance, volumetric modelling and safety automation. The report argues that rapid advances in vehicle control systems and precision machine automation are shifting the industry’s focus from autonomous machines towards more integrated mine-wide operations.

Academic research points in the same direction. In their review of 68 studies examining AI-driven digital twins in mining, Shouki A. Ebad and colleagues argue that existing research remains “scattered across isolated applications rather than unified frameworks”, limiting the industry’s ability to coordinate decisions across whole mining operations. They identify data integration, interoperability and real-time synchronisation as the principal technical barriers to wider deployment.

Lelio Di Martino, general manager of cognitive critical operations at Nokia, echoes that the industry has reached “an inflexion point” where the limiting factor is no longer the capability of individual autonomous machines, but the ability of entire mining operations to make coordinated decisions. 

Although individual systems have become highly capable at the tasks they were designed to perform, he notes that “the operation is not a single system". Operational knowledge remains disctributed across systems designed to solve individual problems rather than optimise the whole operation. 

The next opportunity, he argues, lies in enabling all operations to function as part of a common decision-making process.

That thinking underpins Nokia’s broader vision for the future of autonomous mining. Rather than replacing existing autonomous equipment, Di Martino envisages an intelligence layer capable of combining information from communications networks, sensors, edge computing and operational software into a continuously updated operational picture.

As he argues, industrial organisations have become highly connected without necessarily becoming intelligent, because operational knowledge remains distributed across systems designed to solve individual problems rather than optimise the wider operation.

Di Martino traces its origins to visits to operating mines by researchers from Nokia Bell Labs in 2022. Rather than demonstrating technology, the researchers spent time observing how experienced supervisors made operational decisions.

Those decisions, he recalls, often relied on practical knowledge accumulated over decades rather than information generated by digital systems. 

One example involved deciding when water trucks should be deployed to suppress dust. Another concerned the scheduling of fuel trucks, where several haul trucks could end up waiting simultaneously because refuelling decisions were being made independently of wider production priorities. 

The researchers concluded that no individual software application possessed enough information to optimise those processes.

Understanding vehicle fuel levels alone was insufficient. Just monitoring the refuelling station was equally inadequate. Only by combining information from multiple operational systems could better decisions be made. 

That experience fundamentally changed the scope of the project. 

“Network is only the smallest component of this,” Di Martino says, describing communications infrastructure instead as “the glue” linking operational intelligence. 

Rather than focusing solely on faster connectivity, the research evolved towards understanding how AI could synthesise information already available across the mine.

The practical implications become clearer when considering autonomous drones, says Di Martino. 

Many mines already deploy drones equipped with light detection and ranging and high-resolution cameras to survey pit walls, monitor slope stability and calculate stockpile volumes. However, those surveys often remain separate from day-to-day operational decision-making. 

By the time survey data has been processed, interpreted and distributed, blasting may already have altered the mine. 

Instead, Di Martino envisages information flowing immediately into a shared operational intelligence layer. 

Fresh drone surveys identifying approximately 150,000t of newly blasted rock could automatically update the mine’s digital representation. Dispatch systems could immediately calculate the number of haul trucks required, identify where those vehicles were currently operating and recommend revised haul routes based on the latest terrain.

The objective is not simply to automate trucks or drones individually but to enable one autonomous system to influence another without requiring manual intervention. 

That mirrors broader developments in digital twin research. Ebad and colleagues describe digital twins as creating a continuous feedback loop that enables real-time monitoring, predictive analysis and decision support, evolving beyond traditional simulations towards operational optimisation. 

Di Martino believes mining operations are moving towards that same model, enabling information generated in one part of the operation to influence decisions elsewhere through continuously shared operational intelligence.

Another important aspect of Di Martino’s vision is where those decisions occur. Di Martino uses driver fatigue monitoring as an illustration. 

Conventionally, video captured inside a haul truck may be transmitted to a remote control centre, where operators determine whether a warning represents genuine fatigue or a false alarm.

Instead, Di Martino envisages AI inference being performed directly on edge computing hardware installed within the vehicle, reducing response times from minutes to milliseconds while reducing network traffic. 

The same principle extends to predictive maintenance. Rather than analysing individual vehicles in isolation, Di Martino describes comparing vibration patterns across multiple haul trucks to distinguish between mechanical faults, deteriorating haul roads and changes in loading practices.

A continuously updated digital twin becomes less a visual representation of the mine and more an operational model capable of identifying emerging problems before they affect production. 

That ambition remains technically challenging, however Ebad and colleagues identify data integration, interoperability and real-time synchronisation as the main barriers to AI-driven digital twins, suggesting that connecting autonomous systems may prove as difficult as developing them. 

Likewise, GlobalData’s assessment of innovation trends indicates that AI-enabled fleet orchestration, integrated worksite management and sensor fusion are becoming increasingly important areas of development alongside the deployment of autonomous vehicles themselves.  

Whether such an approach can deliver on its promise will depend on how effectively mining companies overcome those technical challenges across heterogeneous operating environments.

Ebad and colleagues describe AI-driven digital twins as “a fundamental shift in mining engineering, where data is no longer a byproduct but a core asset for real-time safety and environmental optimisation”. 

Mining’s first automation revolution focused on making individual machines autonomous. The next, Di Martino argues, will be judged by how effectively those machines work together.