Sandvik AutoMine Underground Mining
by Nick Clark | Published April 25, 2026
Sandvik AutoMine is the most commercially mature autonomous-equipment platform in underground hard-rock mining, deployed at production scale from Codelco's El Teniente in Chile to Newcrest's Cadia in Australia. The architectural primitive AutoMine still synthesizes ad-hoc — a graduated, machine-checkable commitment ladder for tramming, loading, and dumping cycles in GNSS-denied drifts — is exactly what governed actuation provides.
Vendor and Product Reality
Sandvik Mining and Rock Solutions ships AutoMine as a tiered product family — AutoMine Lite for single-machine teleremote operation, AutoMine Fleet for multi-loader and multi-truck coordination, and AutoMine Concept for surface and tele-operated drilling — running on top of Sandvik's TH and LH series underground trucks and loaders. The platform integrates with the OptiMine production-management suite for shift planning, KPI capture, and route compliance, and it interoperates with third-party fleet-management systems through a published OPC-UA boundary. Major reference deployments include Codelco El Teniente (block-cave production), Newcrest Cadia East (sublevel-cave), Resolute Syama (sublevel-cave with full electric haulage), and LKAB's Kiruna (the largest underground iron-ore mine in the world).
Underground autonomy looks nothing like surface autonomy. There is no GNSS, lighting is artificial and dust-laden, drift walls are mapped by Sandvik's onboard 2D and 3D LiDAR plus inertial dead-reckoning, and traffic is mixed: autonomous LH621i loaders share haul drifts with manned support equipment during shift changes. AutoMine handles this with a production-area concept — a geofenced, isolated zone with light curtains, RFID-tagged interlocks, and a supervisory production controller — inside which autonomous machines operate without human entry, and a teleremote envelope outside it.
The commercial reality is that Sandvik competes head-to-head with Epiroc's Mobilaris and Scooptram Automation, and increasingly with Caterpillar's MineStar Command for Underground. The differentiator is fleet density: AutoMine sites routinely run six to twelve autonomous machines on a single production controller, with throughput numbers (tonnes per hour, availability, mean time between operator interventions) that customers benchmark quarterly against manned baselines.
The Architectural Gap
AutoMine's commitment model today is implicit. A loader's tram-load-tram-dump cycle is encoded as a sequence of waypoints and bucket-actuation primitives, with mission abort handled by a hard-coded fallback: stop in place, raise the bucket, illuminate, and wait for human resolution. There is no graduated response between "execute" and "abort," which means routine perturbations — a piece of muck on the drift floor, a slight LiDAR pose drift, a transient communications dropout to the production controller — escalate into full mission stops that cost real shift minutes.
The gap is sharpest at production-area boundaries and at draw-point loading. When a loader approaches a draw point, it must commit to a bucket trajectory based on a fragmentation model that may be hours stale; if the muck pile has shifted, the bucket strategy needs to degrade smoothly rather than terminate. When a truck queues at a tip, it must commit to a reverse-and-dump maneuver that depends on the geometry of the ore pass and on the tipping cycle of the truck ahead — and a stale or partial sensor view should produce a deferred commitment, not a hard abort.
Underground regulators — Western Australia's DMIRS, Chile's Sernageomin, MSHA in the United States — increasingly require demonstrated post-actuation verification: evidence that an autonomous machine actually executed the commitment its supervisory system authorized. AutoMine produces this evidence today through forensic log reconstruction. That is sufficient for incident investigation but inadequate for continuous compliance, and it is becoming a procurement obstacle as customers move to risk-based assurance frameworks aligned with EMESRT and ISO 17757.
What the AQ Primitive Provides
Governed actuation supplies AutoMine with a typed commitment ladder — continue, defer, refuse, partial — that wraps every actuation surface on a loader or truck: drive, articulation, bucket, tipping cylinder, brake, and lighting. Each commitment carries an explicit precondition envelope (pose uncertainty, communications latency, neighbor-machine reservations, production-area state) and a verifier predicate that closes the loop within a bounded time window after actuation. The primitive is indifferent to whether the supervisory decision came from the on-board planner, the production controller, or a teleremote operator.
In a draw-point loading cycle, the commitment ladder allows the bucket to commit to a partial fill if the fragmentation model disagrees with the live LiDAR view by more than a configured threshold, rather than aborting. In tramming, a transient pose-uncertainty spike produces a deferred commitment — the loader holds its current trajectory, reduces speed, and waits for the next pose update — instead of a full stop. In tip-and-dump, the truck refuses a tipping commitment if the ore-pass approach geometry violates its precondition, and the refusal is structured: the supervisory system receives a typed rationale rather than a generic fault code.
Harm minimization in underground mining is geometric and energetic. A degraded commitment must leave the machine in a state that does not block the haul drift, does not foul the ventilation circuit, and does not commit the battery (on electric LH518B loaders) below the reserve required to clear the production area under remote control. The primitive's reachable-safe-set guarantee captures these constraints as part of the commitment lattice, so a partial or refused commitment is provably recoverable without a rescue sortie.
Composition Pathway
Integration sits at the OPC-UA boundary that AutoMine already exposes between the on-machine controller and the production controller. Existing waypoint and actuation messages are wrapped as commitment-tagged variants; legacy clients continue to see the same surface, while commitment-aware clients see the full ladder, the precondition envelope, and the verifier outcome. Sandvik's existing safety-certified hardware — the on-machine safety PLC, the access-control interlocks, the e-stop network — remains the authoritative safety layer, and the primitive defers to it without contest.
Mesh-coordinated tramming becomes substantially cleaner. When two autonomous loaders converge in a haul drift, today's arbitration is a centralized lock managed by the production controller. With governed actuation, each loader publishes its current and intended commitments, and a distributed reservation protocol resolves the convergence locally with the production controller as a witness rather than as a synchronous arbiter. This reduces communications-link sensitivity, which is the dominant source of unplanned stops in deep-mine deployments where leaky-feeder or 5G coverage is uneven.
Verifier libraries are the natural site for Sandvik-specific intellectual property. The fragmentation-aware bucket verifier, the ore-pass tip verifier, and the articulation-against-drift-wall verifier all encode Sandvik's geomechanical and machine-dynamics knowledge. Adaptive Query's reference primitive ships with generic verifiers; Sandvik supplies the production-grade ones, and the licensing carve-out preserves that boundary.
Commercial and Licensing Implication
For Sandvik, the primitive converts a procurement-cycle pain point into a procurement-cycle advantage. Customers running risk-based assurance frameworks — Rio Tinto, BHP, Fortescue, Vale — increasingly demand auditable evidence of post-actuation verification as a condition of approving autonomous deployment in new production areas. AutoMine with governed actuation produces that evidence as a designed-in artifact, shortening the path from FAT (factory acceptance) through SAT (site acceptance) to first autonomous tonnes by a measurable margin.
Licensing is non-exclusive across mining-equipment OEMs, and Adaptive Query's expectation is that Epiroc, Caterpillar, and Komatsu adopt the same primitive. The shared commitment vocabulary is what allows mixed-fleet sites — common in brownfield expansions — to interoperate at the supervisory layer without a bespoke integration per OEM. Sandvik's competitive advantage is fleet density and verifier maturity; the primitive is the shared substrate that makes that advantage legible to customers and to regulators.
Insurance, finance, and ESG reporting layers extend the same logic. Mining-equipment financiers and operational-loss underwriters increasingly price autonomy risk against the auditability of the supervisory record; a typed commitment lattice with verifier outcomes converts a category of unstructured operational losses into structured, replayable events. ESG reporting under the GRI Mining and Metals supplement and ICMM principles benefits from the same auditability, because the commitment lattice captures not only what the machine did but what it refused to do and why — a level of operational transparency that manned operations have never been required to produce and that autonomous operations are now expected to.
