⚡ electrical engineering

Inertial Magnetic SLAM Systems Using Low-Cost Sensors

This paper proposes and evaluates two low-cost, non-visual Inertial Magnetic SLAM systems that utilize IMUs, magnetometer arrays, and barometers to achieve bounded-error 3D positioning and mapping in challenging environments, with experimental results demonstrating that a tightly coupled approach outperforms a loosely coupled one.

Original authors: Chuan Huang, Gustaf Hendeby, Isaac Skog

Published 2026-04-14

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Chuan Huang, Gustaf Hendeby, Isaac Skog

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 you are walking through a massive, windowless maze in total darkness. You have no GPS (because you're underground or inside a building), no map, and you can't see anything. How do you know where you are?

This is the problem Inertial Magnetic SLAM tries to solve. The paper you shared presents a new way to navigate these "blind" spaces using only cheap sensors found in smartphones and a special trick: reading the invisible magnetic fingerprints of a building.

Here is the breakdown in simple terms, using some fun analogies.

1. The Problem: The "Drifting" Compass

Most indoor navigation systems rely on IMUs (Inertial Measurement Units), which are like a blindfolded person trying to walk in a straight line. They count their steps and guess their direction.

The Flaw: If you take 100 steps, you might think you walked 100 meters. But if you are off by just a tiny fraction of a degree, after 100 meters, you could be 10 meters off course. This is called drift. Over time, the "blindfolded person" gets completely lost.

To fix this, old systems used cameras (Visual SLAM) or laser scanners (LiDAR).

The Flaw: Cameras fail in the dark or if the walls are blank. Laser scanners are expensive and bulky.

2. The Solution: The Building's "Magnetic Fingerprint"

Every building is full of steel beams, pipes, and electronics. These create tiny, unique distortions in the Earth's magnetic field.

The Analogy: Imagine the building is a giant, invisible magnetic fingerprint. Even though the magnetic field looks smooth from far away, if you zoom in, every corner, hallway, and room has a unique "magnetic texture."
The Sensor: The researchers built a board with 30 magnetometers (tiny magnetic sensors) and one IMU. Instead of just one sensor, they have an array (a team of sensors). This is like having 30 people holding hands, feeling the texture of the wall all at once, rather than just one person feeling it with a single finger.

3. The Two New Systems: "Loose" vs. "Tight"

The paper proposes two ways to combine the "step-counting" (IMU) with the "magnetic fingerprint reading" (Magnetometers).

A. The Loosely Coupled System (The Relay Race)

How it works: Imagine a relay race.
1. Runner 1 (The IMU): Runs fast, counting steps and guessing the path. Every second, they hand a "position estimate" to Runner 2.
2. Runner 2 (The SLAM): Takes that estimate, looks at the magnetic map, and says, "Wait, you're actually here, not there." They correct the path and hand the new, corrected position back.
The Issue: Because Runner 1 is guessing the path before Runner 2 checks it, any small mistake Runner 1 makes gets passed along. It's like a game of "Telephone" where the message gets slightly distorted at every hand-off.

B. The Tightly Coupled System (The Huddle)

How it works: Imagine a football team in a huddle. Everyone (the IMU, the 30 magnetometers, and the barometer) talks to each other simultaneously.
- They don't pass a message back and forth. They pool all their data at the exact same moment to figure out the position.
- The IMU says, "I think I moved forward."
- The Magnetometers say, "But the magnetic field here looks like the one near the elevator, not the stairs."
- The Barometer says, "And I feel like we are on the 3rd floor."
- Result: They solve the puzzle together instantly.
The Result: This system is much smarter. It catches errors immediately before they pile up. The paper shows this "Huddle" approach is significantly more accurate than the "Relay Race."

4. The "Secret Sauce": The Barometer

The system also uses a barometer (an air pressure sensor).

The Analogy: In a multi-story building, the magnetic field might look similar on the 2nd and 3rd floors. The barometer acts like an elevator detector. It tells the system, "Hey, the air pressure dropped, we just went up a floor!" This prevents the system from getting confused about which floor it is on.

5. The Results: Walking the Walk

The researchers tested this in a real building with multiple floors, long corridors, and spiral staircases.

The Performance: Over a walk of about 200 meters (roughly two football fields), the best system (Tightly Coupled) was only off by less than 5 meters.
Why it matters: That is a huge improvement for a system using only cheap sensors. It means a firefighter could walk into a burning building, map the inside, and know exactly where they are without needing GPS or expensive lasers.

Summary

Think of this technology as giving a robot (or a first responder) a superpower: the ability to "feel" the invisible magnetic shape of a building to know exactly where they are, even in total darkness.

Old way: Guessing steps (Drifts away).
Visual way: Needs light and cameras (Fails in smoke/dark).
This new way: Uses a team of 30 magnetic sensors to "read" the building's invisible magnetic map, correcting the step-counting in real-time.

The paper proves that you don't need expensive, high-tech gear to navigate complex buildings; you just need a clever algorithm and a board full of cheap magnetic sensors.

1. Problem Statement

Global Navigation Satellite Systems (GNSS) are unreliable or unavailable in indoor and subsurface environments, posing significant challenges for positioning and navigation. While Visual and LiDAR SLAM systems are common alternatives, they suffer from limitations such as sensitivity to lighting conditions, high costs, and dependence on geometric features.

Existing magnetic field-based SLAM systems offer a cost-effective solution using ubiquitous ferromagnetic materials in buildings as environmental features. However, state-of-the-art magnetic SLAM methods typically rely on low-drift odometry (from wheel encoders, visual odometry, or pedestrian dead-reckoning) to constrain motion. Furthermore, many systems assume 2D planar motion or use single magnetometers, which provide limited spatial information.

The core research questions addressed are:

Can a full 3D Inertial Magnetic SLAM (IM-SLAM) system be built using only low-cost sensors (IMU, barometer, and a magnetometer array) without external odometry?
How does a tightly coupled IM-SLAM architecture compare to a loosely coupled one in terms of accuracy and robustness?
How do sensor quality (IMU) and auxiliary sensors (barometer) impact system performance?

2. Methodology

The authors propose two distinct system architectures that integrate a Magnetic-Field-Aided Inertial Navigation System (MAINS) with a Magnetic Field SLAM framework. Both systems utilize a custom sensor board containing a low-cost IMU, a barometer, and an array of 30 magnetometers.

A. Sensor Models

Local Model: Uses a first-degree polynomial model to approximate magnetic field variations within the small area covered by the magnetometer array. This allows for "odometry" based on magnetic field gradients.
Global Model: Uses an approximated Gaussian Process (GP) model to map the magnetic field across the entire environment (building scale). This provides global consistency and loop closure.

B. System Architectures

Loosely Coupled IM-SLAM:
- Structure: A cascaded system operating on two time scales.
- Process: The modified MAINS runs at the high IMU rate (100 Hz) to compute relative odometry (position and orientation changes) between time steps. This odometry is then fed as a measurement into a lower-rate (downsampled) Magnetic Field SLAM filter.
- Limitation: The odometry data generated by the MAINS is temporally correlated (successive estimates overlap), which violates the independence assumptions often made in standard SLAM filters, potentially degrading performance.
Tightly Coupled IM-SLAM:
- Structure: A unified framework where all states are estimated simultaneously in a single filter.
- State Vector: Includes inertial navigation states (position, velocity, orientation, biases) and coefficients for both the local polynomial model and the global GP model.
- Measurement Strategy: The system alternates between two measurement equations:
  - High Frequency: Uses the efficient local polynomial model (from MAINS) for most updates.
  - Periodic Updates: Every $D$ samples, it uses the complex global GP model to correct drift and enforce global consistency.
- Advantage: By estimating navigation states and map parameters jointly, it avoids the intermediate odometry computation step, thereby resolving the temporal correlation issues inherent in the loosely coupled approach.

3. Key Contributions

Novel System Design: Proposes the first 3D IM-SLAM systems that rely solely on low-cost sensors (IMU, barometer, 30-magnetometer array) without requiring visual or wheel odometry.
Architectural Comparison: Provides a rigorous comparison between loosely and tightly coupled architectures, demonstrating that the tightly coupled approach yields superior accuracy by handling state estimation in a single step.
Ablation Studies:
- Barometer: Analyzed the impact of barometric aiding for height stabilization.
- Sensor Quality: Evaluated system robustness by degrading IMU quality (simulating lower-cost sensors).
- Initialization: Determined the required duration of initial pose aiding (e.g., GPS or motion capture) for stable bias estimation.
Open Source: The datasets and algorithms are made publicly available to ensure reproducibility.

4. Experimental Results

The systems were evaluated in real-world indoor environments with multi-floor structures using three trajectory types: Long Corridor, Corridor (square loops), and Spiral Staircase.

Accuracy:
- The Tightly Coupled IM-SLAM consistently outperformed both the MAINS-only system and the Loosely Coupled IM-SLAM.
- Performance Metric: The tightly coupled system achieved a final positioning error of less than 5 meters over a ~200-meter trajectory (approx. 2.5% error rate).
- In specific scenarios (e.g., Long Corridor-2 with overlapping paths), the tightly coupled system correctly identified loop closures where the loosely coupled system failed due to magnetic field ambiguities.
Ablation Findings:
- Barometer: While the system is not heavily reliant on barometer data for horizontal accuracy, the barometer is crucial for vertical stability and distinguishing between floors in multi-story buildings.
- IMU Quality: The tightly coupled system demonstrated remarkable robustness. Even when IMU data quality was degraded by averaging down to a single sensor (simulating a factor of 32 reduction in sensor quality), the system maintained acceptable accuracy.
- Initialization: A pose aiding duration of at least 20 seconds at the start of the trajectory is recommended to stabilize IMU bias estimates effectively.
Trajectory Smoothing: The authors introduced a "pseudo-measurement" technique to smooth the jerky segments in the tightly coupled system's trajectory, resulting in smoother paths without sacrificing accuracy.

5. Significance and Applications

Feasibility of Low-Cost 3D SLAM: The paper proves that high-accuracy 3D indoor positioning is achievable without expensive LiDAR or cameras, using only commodity-grade sensors.
Emergency Response: The primary application identified is first responder positioning (e.g., firefighters, mine rescue). In these scenarios, GNSS is unavailable, and visual sensors may fail due to smoke, darkness, or dust. Magnetic fields are ubiquitous and unaffected by these conditions.
Scalability: The use of a magnetometer array allows for "magnetic odometry," providing motion constraints that are essential for SLAM in feature-poor environments where visual odometry fails.

In conclusion, this work establishes a robust framework for Inertial Magnetic SLAM, demonstrating that a tightly coupled approach using low-cost sensors can achieve meter-level accuracy in complex 3D indoor environments, offering a viable solution for GNSS-denied navigation.