🔬 applied physics

From Rubble Simulation to Active Magnetic Mapping: Quantum Sensing for Disaster Response

This paper proposes and validates a drone-based quantum magnetometry pipeline that combines Unreal Engine rubble simulations with Bayesian active sampling to effectively reconstruct the magnetic structure of collapsed buildings for survivor detection within the critical 72-hour window.

Original authors: Samuel Tovey, Stefan Prestel, Hiroshi Yamauchi

Published 2026-06-25

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Samuel Tovey, Stefan Prestel, Hiroshi Yamauchi

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a building has collapsed, like a giant tower of blocks knocked over by an earthquake. Inside the pile of rubble, there might be survivors, but it's incredibly hard to see where they are. Current tools (like cameras or heat sensors) can only peek through the cracks, giving us a blurry picture of what's underneath.

This paper proposes a new way to "see" the rubble using drones equipped with super-sensitive quantum magnets. Here is how they tested it and what they found, explained simply:

1. The Problem: The "Invisible" Rubble

When a building made of steel-reinforced concrete falls, the steel inside doesn't disappear. It just gets buried. Because steel interacts with the Earth's magnetic field, the pile of rubble actually creates a faint, invisible magnetic "shadow" or pattern.

The challenge is that this magnetic signal is incredibly weak—so weak it's like trying to hear a whisper from a mile away. The researchers wanted to know: Can a drone flying above the rubble detect this whisper, and can we turn those faint whispers into a clear map?

2. The Simulation: A Digital "What-If"

Since they couldn't blow up a real building to test this, they built a digital twin of a parking garage in a video game engine (Unreal Engine).

They simulated an earthquake that shattered the concrete roof and pillars.
They programmed the computer to calculate how the broken steel pieces would interact with the Earth's magnetic field.
The Result: They found that even from 1 to 2 meters above the pile, the drone could detect a magnetic pattern. It wasn't a perfect picture, but the "outline" of the standing pillars and large roof chunks was visible. The signal was tiny (trillionths of a Tesla), but modern quantum sensors are sensitive enough to catch it.

3. The Solution: The "Smart Drone" Strategy

Detecting the signal is only half the battle. If a drone flies in a boring, back-and-forth pattern (like a lawnmower), it wastes battery and time. The researchers tested three ways for the drone to fly:

The Lawnmower: Flying straight lines back and forth. (Reliable but slow).
The Sine Wave: Flying in a smooth, wavy pattern. (Faster, but sometimes misses the "sweet spots").
The "Smart" Explorer (Bayesian Active Sampling): This is the star of the show. The drone flies to a spot, checks the magnetic field, and then uses a smart algorithm to ask, "Where is the most confusing part of the map that I haven't seen yet?" It then flies there to learn more.

The Finding: The "Smart Explorer" was the winner. It built a clear picture of the rubble's structure using only about 100 samples (flight points). The other methods needed hundreds more points to get the same clarity.

4. The Hardware: How Many Sensors?

The drone carried a small array of magnetic sensors. The team tested carrying 1, 2, 3, or 4 sensors.

1 Sensor: Too shaky; the map was unreliable.
4 Sensors: Didn't add much extra value compared to 3.
3 Sensors: This was the "Goldilocks" zone. It offered the best balance between getting a sharp image and keeping the drone light enough to fly.

5. The Big Picture

The paper concludes that this technology is feasible right now.

The Physics: The magnetic signals from collapsed steel buildings are strong enough to be detected by current quantum sensors from a safe distance.
The Math: Using a smart algorithm, a drone can fly just a few hundred times over a disaster site and reconstruct a 3D map of the hidden structure.

In short: By combining super-sensitive quantum magnets with a "smart" flight path, drones could potentially create a clear map of a collapsed building's interior without needing to dig through the rubble first. This could help rescue teams find safe paths or locate empty spaces (voids) where survivors might be hiding.

Note: The paper focuses entirely on the simulation and the ability to map the structure. It does not claim to have found survivors or tested this on a real disaster site yet; it proves the concept works in a computer simulation.

Technical Summary: From Rubble Simulation to Active Magnetic Mapping

Problem Statement
Locating survivors within the first 72 hours of a building collapse is a critical global challenge, yet existing sensing modalities (audio, thermal, ground-penetrating radar, etc.) provide only partial information regarding the internal structure of rubble. While Magnetic Simultaneous Localization and Mapping (MagSLAM) has been explored for GPS-denied positioning, the complementary task of reconstructing the underlying magnetic field of a collapsed structure from a known platform pose remains under-addressed. Classical magnetometers limit the resolution of such maps. This paper investigates whether stable quantum sensors, specifically Nitrogen-Vacancy (NV) centers in diamond, can enable drone-based structural analysis and void detection by sensing subtle magnetic fluctuations in steel-reinforced concrete.

Methodology
The authors developed an integrated simulation pipeline spanning three stages, executed on an NVIDIA H100 GPU using JAX:

Rubble Physics Simulation: Using Unreal Engine's Chaos physics engine, the team modeled the collapse of a 28m × 28m × 5m steel-reinforced concrete parking garage. The simulation utilized Voronoi tessellation to fracture the mesh under a lateral impulse. Crucially, the model grouped fractures into nested blocks to simulate the binding action of rebar, preventing unrealistic total disaggregation of concrete slabs.
Magnetic Field Modeling: The rubble mesh was converted into a ground-truth magnetic field using a per-triangle dipole approximation. Each triangle was treated as a thin slab with an induced magnetic moment ( $m_i = A_i \chi B_{earth}$ ) driven by the local Earth's magnetic field. The total field at any point was calculated as the sum of dipoles. Parameters were set for a Tokyo deployment, using a conservative bulk susceptibility ( $\mu_r = 30$ ) for reinforced concrete.
Reconstruction and Sampling: An autonomous mapping backend utilized Gaussian Process Regression (GPR) with a Radial Basis Function (RBF) kernel to reconstruct continuous magnetic topologies from sparse sensor data. The system compared three flight strategies: a deterministic "lawnmower" path, a sine-wave sweep, and a Bayesian active sampling strategy (using an Upper Confidence Bound acquisition function) to optimize the trade-off between exploration and exploitation.

Key Contributions and Results

Physical Observability: The simulation established that meaningful magnetic structure is recoverable from steel-reinforced concrete rubble. While fields near the source surface range from 30 pT to 3 µT, the structurally informative contrast at UAV operating altitudes (1–2 meters above the roofline) falls within the sub-pT to sub-nT range. This range is well within the sensitivity envelope of modern room-temperature quantum magnetometers. The study found that intact pillars and roof panels dominate the field signature, while detached fragments contribute minimally.
Sensor Array Optimization: The study evaluated array sizes ( $N \in \{1, 2, 3, 4\}$ ). Results indicated that a three-sensor array offers the optimal trade-off between gradient resolution and UAV payload constraints. While increasing sensors generally improved reconstruction quality, gains plateaued or regressed at $N=4$ for fragmented collapse scenarios, particularly under phase-sensitive trajectories.
Sampling Efficiency: The Bayesian active sampling strategy significantly outperformed deterministic methods. It reached peak structural correlation (measured by Pearson correlation against ground truth) in approximately 100 samples, whereas deterministic strategies required significantly more data to achieve similar fidelity. The active sampler effectively identified the gross structure of the rubble with high fidelity from sparse multi-sensor samples.
Sensor Spacing: Reconstruction quality improved monotonically with sensor inter-element spacing (from 5 cm to 2 m), confirming the role of the array as a local gradient estimator. However, the paper notes the practical difficulty of mounting meter-scale footprints on small UAVs.

Significance and Claims
The paper claims to bridge the physical and algorithmic sides of disaster response sensing. It demonstrates that:

Feasibility: The magnetic signatures of collapsed steel-reinforced structures are strong enough to be detected by quantum-grade sensors at safe standoff distances, providing a new modality for structural analysis.
Efficiency: A combination of a three-sensor quantum array and Bayesian active sampling can reconstruct the gross structural map of a collapse in roughly 100 samples, a regime highly relevant for time-critical search-and-rescue operations.
Application Potential: The reconstructed field maps could serve as actionable inputs for void detection (identifying non-magnetic gaps where survivors might be located) and structural integrity assessment to guide safe debris removal.

The authors maintain a modest tone, acknowledging that the current dipole approximation neglects remnant magnetization and near-field coupling, which may require surface-integral formulations for ground-based robots. They conclude that while the pipeline is simulated, both the sensing requirements and the reconstruction algorithms are within reach of current platforms.