How We Turn Reality Capture into Actionable Mine Data
Estimated reading time: 4 minutes
When people hear that we use drones to map underground mines, they usually assume the hardest part of the job is flying the aircraft. It’s an understandable assumption — underground mines have no GPS, poor lighting, dust-filled air, and passages that narrow unexpectedly into large voids. Every flight has to be planned around operational schedules, ventilation, battery limits, and safety.
But here’s what surprises most people: the drone flight is usually the shortest part of the entire project. A mission might take twenty minutes. Turning that data into engineering information can take several days. The drone isn’t the product — it’s the tool used to collect reality. What engineers, surveyors, and planners actually need are deliverables: CAD drawings, watertight meshes for volumetric calculations, inspection models, and georeferenced datasets that slot into existing survey workflows. None of that exists the moment the drone lands. It has to be engineered, and this is where we excel!
It Starts With a Question, Not a Drone
At Vision AI every project begins with the engineering problem, not the aircraft. “We need to calculate the volume of this stope,” “we need drawings before rehabilitation begins,” and “we need to inspect an area that’s unsafe to enter” might all use the same drone — but they demand completely different capture strategies, levels of overlap, and coordinate control.
That’s why, before batteries are even charged, we work through what decisions the deliverables need to support, what accuracy is required, and which software environment the client will use. We don’t try to capture as much data as possible — we capture the right data for the engineering outcome.
Planning Happens Long Before the Propellers Spin
Underground surveying leaves little room for improvisation. There’s no flying higher to avoid an obstacle, no GPS for positional awareness, and a missed section can mean a costly return trip if access schedules change or production resumes. So our first objective is simple: capture everything needed the first time. That means reviewing existing mine plans, coordinating with surveyors and safety personnel, and accepting that assumptions made in the office rarely survive first contact with the underground environment — drives get extended, ground conditions shift, and equipment appears where the plans say there’s open space.
Capturing Reality Without GPS
Underground, drones rely on SLAM — Simultaneous Localisation and Mapping — to solve two problems at once: where am I, and what does the environment around me look like? LiDAR, inertial sensors, and visual data work together continuously to answer both. But SLAM isn’t magic; its accuracy depends entirely on how well the flight is executed. That’s why speed is rarely the goal underground. Flying too fast reduces scan density and overlap, and — much like skimming every third page of a book — leaves gaps that only become obvious once the flight is over and it’s too late to fix.
The Real Work Starts When the Drone Lands
Raw LiDAR data isn’t a deliverable. It can’t be imported into planning software or used for reconciliation until it’s been reconstructed, cleaned, and validated — a process that often takes longer than the flight itself. The trajectory is refined, the point cloud is assembled, and then comes one of the most underrated steps: cleaning. Dust, reflections, and stray returns all get recorded alongside genuine geology, and separating one from the other takes engineering judgement, not just software.
Registration and validation matter just as much. A model can look flawless and still be wrong — we once had a reconstruction that looked perfect in every respect, until someone opened it in GIS software and found the mine sitting thousands of kilometres from where it should be. The scan hadn’t failed; a coordinate reference mismatch had. It was caught during QA rather than after delivery, and it’s a lesson we now build into every project: a visually correct model isn’t necessarily a spatially correct one.
Only once a dataset clears these checks does it branch into the products clients actually use: georeferenced point clouds, OBJ meshes, DXF drawings, cross-sections, volumetric calculations, and inspection reports.
One Survey, Many Stakeholders
A single underground flight rarely serves just one purpose. Surveyors use it to verify geometry and build a permanent digital record. Planners need structured drawings that drop straight into CAD. Production engineers rely on validated surfaces for volumetric reconciliation, where even small gaps in coverage can shift the numbers. Geotechnical engineers use it to assess hazardous ground without setting foot in it. And operations managers just want clarity — a clear visual they can bring into tomorrow’s planning meeting. One dataset, five very different jobs.
What We’ve Learned
Better sensors and faster software haven’t made underground surveying simple — they’ve just removed some of the practical barriers. Judgement calls like whether an excavation has been fully documented, or whether a surface is suitable for volumetric reconciliation, still belong to experienced engineers. Most project problems don’t start with bad data in the field; they start with unclear objectives before the drone ever leaves the ground. And quality assurance works best when it’s built into every stage, not bolted on at the end.
Reality capture is increasingly becoming the foundation of the digital mine — a growing, evolving record that supports far more than the survey it was originally collected for. But the drone was never really the story. The story is everything that happens after it lands: the planning, the processing, the validation, and the deliverables that help people make better decisions underground.
This article is the first in Engineering the Underground, a technical blog series by Vision AI exploring the workflows and engineering practices behind modern underground reality capture.