🎓 Lesson 19
D5
El Teniente Underground Localization Lab
El Teniente Underground Localization Lab teaches how to precisely locate and track autonomous haul trucks underground using sensors, radios, and 3D maps so they can move safely and efficiently in low-visibility tunnels.
🎯 Learning Objectives
- ✓ Calculate achievable localization uncertainty given UWB anchor geometry and noise floor
- ✓ Design a minimal UWB anchor layout for a 6-m-wide, 500-m-long decline using coverage and dilution-of-precision (DOP) constraints
- ✓ Analyze IMU drift accumulation over 5-minute dead-reckoning intervals under typical underground vibration spectra
- ✓ Explain trade-offs between map-based LiDAR SLAM and infrastructure-dependent UWB localization in terms of update rate, accuracy, and maintenance overhead
- ✓ Apply ISO 17757:2019 safety thresholds to validate end-to-end localization integrity for SIL-2 AHS control loops
📖 Why This Matters
At El Teniente—the world’s largest underground copper mine—over 120 autonomous Komatsu HD785 haul trucks operate daily in tunnels up to 1.2 km below surface, where GPS is unavailable and RF propagation is highly unpredictable. Precise, reliable localization isn’t just about navigation—it’s the foundational safety layer preventing collisions, enabling collision avoidance, and ensuring payloads are delivered to correct drawpoints. Failure here cascades into production stoppages, equipment damage, and regulatory noncompliance. This lab bridges theory with the exact hardware, constraints, and data from El Teniente’s operational AHS fleet.
📘 Core Principles
Underground localization relies on three complementary layers: (1) Infrastructure-based ranging (e.g., UWB anchors installed on tunnel walls) provides absolute position but suffers from line-of-sight loss and anchor misalignment; (2) Inertial navigation (IMU + wheel odometry) enables short-term continuity but drifts due to bias instability and vibration-induced errors; (3) Map-aided perception (LiDAR SLAM or vision-based feature matching) corrects drift by aligning live sensor data to a georeferenced 3D mine model—but requires sufficient geometric texture and fails in smooth-lined or dust-laden sections. True robustness emerges only when these layers are fused via Kalman or factor-graph filters, with integrity monitoring that detects and isolates faulty measurements per ISO 13849-1 and IEC 61508.
📐 Geometric Dilution of Precision (GDOP) for UWB Anchor Layout
GDOP quantifies how anchor geometry amplifies range measurement errors into position uncertainty. Lower GDOP (<2.5) indicates optimal geometry; >6.0 implies poor coverage and unreliable 3D fixes. Used during anchor placement design to ensure worst-case localization error remains within SIL-2 safety bounds.
GDOP (3D)
GDOP = \sqrt{\text{trace}\left((H^TH)^{-1}\right)}Quantifies sensitivity of position solution to range measurement errors based on anchor geometry.
Variables:
| Symbol | Name | Unit | Description |
|---|---|---|---|
| H | Geometry matrix | dimensionless | 4×3 matrix whose rows are unit vectors from vehicle to each UWB anchor |
| trace | Matrix trace | dimensionless | Sum of diagonal elements of (HᵀH)⁻¹ |
Typical Ranges:
Optimal anchor tetrahedron: 1.0 - 2.5
Linear anchor layout (poor vertical resolution): 4.0 - 12.0
💡 Worked Example
Problem: Four UWB anchors are placed at corners of a 6 m × 6 m horizontal cross-section, spaced every 50 m along a 500 m decline (z-axis). Anchor coordinates: A₁(0,0,0), A₂(6,0,0), A₃(6,6,0), A₄(0,6,0). A truck is at P(3,3,25). Range measurement standard deviation σᵣ = 0.15 m. Calculate GDOP at P.
1.
Step 1: Compute unit line-of-sight vectors from P to each anchor (e.g., u₁ = (A₁−P)/‖A₁−P‖ = (−3,−3,−25)/25.42 ≈ [−0.118, −0.118, −0.979])
2.
Step 2: Assemble geometry matrix H (4×3) with rows = [uₓ, u_y, u_z]; compute HᵀH and invert it
3.
Step 3: GDOP = √[trace((HᵀH)⁻¹)] = √(0.84 + 0.84 + 0.11) ≈ √1.79 ≈ 1.34
Answer:
The GDOP is 1.34, well within the optimal range (<2.5), indicating high geometric strength for this anchor configuration at mid-tunnel elevation.
🏗️ Real-World Application
In Q3 2023, Codelco deployed a UWB+IMU+LiDAR fusion system across El Teniente’s Nueva Explotación decline. Prior to fusion, standalone UWB showed ±0.45 m 2σ horizontal error in straight sections but >±2.1 m in sharp 15° switchbacks due to anchor occlusion. After integrating ZUPT (Zero Velocity Update) during loading pauses and loop-closing LiDAR SLAM against a surveyed 3D mesh (accuracy ±0.08 m), end-to-end 95% horizontal localization uncertainty was reduced to ±0.28 m—even during continuous operation. This met ISO 17757:2019 Annex B requirements for ‘Class 2’ AHS (max 0.3 m lateral error for 95% of time), enabling approval for unattended night-shift hauling.
🔧 Interactive Calculator
🔧 Open Autonomous Haulage Systems Integration Calculator📋 Case Connection
📋 Underground Copper Mine AHS Deployment at Codelco El Teniente
Limited GNSS availability, high dust, and narrow ramps requiring <1.2m lateral accuracy