Autonomous LHDs: AI and 3D LiDAR in Underground Mining

Underground mining is undergoing a massive safety and efficiency revolution through the deployment of autonomous Load-Haul-Dump (LHD) machines. Operating in pitch-black, GPS-denied environments filled with toxic post-blast fumes and heavy particulate dust, these robotic vehicles rely on a combination of 3D LiDAR and artificial intelligence to navigate narrow tunnels and clear blasted ore. This article explores the exact mechanics of how autonomous LHDs map underground drifts, see through dense dust, and execute complex mucking operations without putting human operators in harm's way.

Satellite signals cannot penetrate solid rock, making traditional GPS useless in underground mines. To move through underground tunnels—known as "drifts"—autonomous LHDs utilize a technology called Simultaneous Localization and Mapping (SLAM).

High-speed 3D LiDAR (Light Detection and Ranging) sensors mounted on the machine emit millions of laser pulses per second. These lasers bounce off the uneven rock walls of the drift and return to the scanner, generating a real-time, three-dimensional point cloud map of the environment. The onboard AI compares this live point cloud to a pre-existing digital blueprint of the mine, allowing the LHD to pinpoint its exact location within centimeters and steer safely at high speeds without hitting the tunnel walls.

Seeing Through Toxic Dust with AI Filtering

After a rock face is blasted with explosives, the area is filled with toxic gases (like carbon monoxide and nitrogen dioxide) and thick, blinding silica dust. Historically, mines had to pause operations for hours to ventilate the shafts before human workers could enter. Autonomous LHDs can enter immediately, provided their sensors can overcome the dust.

Standard optical cameras are blinded by heavy dust, and even LiDAR lasers can accidentally bounce off airborne dust particles, creating "false positives" that look like solid obstacles to a computer. AI solves this through advanced spatial and temporal filtering algorithms:

  • Multi-Echo LiDAR: The sensors record multiple returns from a single laser pulse. Part of the light bounces off the dust, while the rest pushes through and hits the solid rock wall behind it.
  • AI Point-Cloud Denoising: Machine learning models analyze the density and movement of the laser reflections. The AI recognizes the chaotic, shifting patterns of a dust cloud and digitally deletes those points from the map, leaving only the solid, stationary geometry of the tunnel and the ore pile.

Automated Mucking: Scooping the Blasted Ore

"Mucking" is the process of scooping up the heavy, fragmented rock (the muck pile) left behind after a blast. While navigating a tunnel only requires following a path, mucking requires the machine to physically interact with an unpredictable, shifting pile of jagged boulders.

When the autonomous LHD approaches the blast zone, its 3D LiDAR scans the muck pile to determine its exact volume, shape, and slope. The AI then takes over the vehicle's hydraulics and drivetrain to execute the scoop:

  1. Target Selection: The AI identifies the optimal entry point at the base of the pile, avoiding oversized boulders that could jam the bucket.
  2. Traction Control: As the LHD rams the bucket into the rock pile, onboard sensors monitor wheel slip, hydraulic pressure, and engine torque. The AI dynamically adjusts the throttle and bucket angle to prevent the tires from spinning out on the wet, uneven floor.
  3. Bucket Fill Optimization: Using proprietary crowding-and-curling motions, the AI wiggles and lifts the bucket through the pile until weight sensors confirm a full load.

Hauling and Dumping the Load

Once the bucket is full, the LHD reverses out of the blast zone. As it backs away, the 3D LiDAR scans the muck pile one last time. The AI updates the mine's central map, calculating exactly how much ore remains and deciding the best angle of attack for its next pass.

The machine then transitions to its hauling phase, navigating back through the drift to an ore pass (a vertical drop chute) or waiting haul truck. Using its LiDAR to align precisely with the dump edge, the LHD tips its bucket, empties the payload, and immediately returns to the toxic blast zone to repeat the cycle. Meanwhile, a single human supervisor sits safely in a ventilated, surface-level control room miles away, overseeing an entire fleet of these autonomous machines.