Fast and Noise-Resilient Magnetic Field Mapping on a Low-Cost UAV Using Gaussian Process Regression

Prince Kuevor, Maani Ghaffari, Ella Atkins, James Cutler

TL;DR

This paper presents a robust framework for creating accurate 3D indoor magnetic field maps using low-cost UAVs. By combining PID controller tuning, physical sensor isolation, and a novel "compromise" Gaussian Process Regression (GPR) model, the authors successfully mitigate vehicle-induced electromagnetic noise and flight-to-flight biases to enable reliable indoor localization.

What Problem Does This Paper Address?

Indoor navigation remains a challenge for robotics because GPS signals are unavailable. While the Earth's magnetic field offers a ubiquitous reference, indoor environments are filled with metallic structures (beams, elevators, pipes) that warp the field. To use these distortions for navigation (a technique called "fingerprinting"), one must first create an accurate map.

UAVs are ideal for this mapping task because they can quickly cover a 3D volume. However, they are "magnetically loud." The high-current motors and Electronic Speed Controllers (ESCs) generate significant electromagnetic interference (EMI). Furthermore, small variations in battery state or hardware initialization can cause the UAV to inject different magnetic biases from one flight to the next, making it difficult to create a consistent, reusable map.

Methods

The authors propose a multi-layered approach to clean up the magnetic data:

  1. Software Noise Reduction: They discovered that high-frequency motor commands—driven by the derivative (D) term in the flight controller’s PID loops—were a major source of magnetic noise. By zeroing out these specific gains ("quiet" gains), they significantly reduced measured variance.
  2. Hardware Isolation: They implemented the standard practice of distancing the magnetometer from the main electronics using a mounting boom (varying distance $S_2$).
  3. Compromise GPR Map: Standard Gaussian Process Regression (GPR) is computationally expensive. The authors introduced a "compromise" model. They train hyperparameters on a large, multi-flight dataset ($n_2$) to capture general trends and flight-to-flight variations. For actual inference, they use a smaller, spatially representative subset of points ($n_1$). This prevents overfitting to a single flight's bias while keeping computation fast.
  4. Consistency Test: They developed a metric to determine if new flight data matches an existing map. If 96% of new observations fall within two standard deviations ($2\sigma$) of the GPR map's prediction, the data is considered "consistent."

Datasets

The researchers collected data at the University of Michigan Robot Fly Lab, an indoor arena equipped with an OptiTrack motion capture system for ground-truth positioning.

  • Training/Validation Sets: They used several flight series (labeled t1 through t6) consisting of "lawnmower" scanning patterns at multiple altitudes.
  • Hardware: A custom quadrotor ("Q1") equipped with a BeagleBone Blue and an RM3100 magnetometer.

Key Insights and Ideas

  • The PID-Magnetic Link: One of the most interesting findings is that flight control software directly impacts sensing quality. Vibrations picked up by the gyroscope are amplified by the PID derivative term, leading to high-frequency current fluctuations that the magnetometer perceives as noise.
  • Spatial Density Limits: The authors found that a spatial density of roughly 0.55m between training observations is the "sweet spot." Increasing density beyond this does not significantly improve map accuracy for indoor environments.
  • Vector vs. Norm Mapping: They empirically proved that creating a specialized map for the magnetic field's magnitude (norm) is unnecessary. Taking the norm of the estimates from a 3-axis vector map yields nearly identical results (within 1nT difference).
  • Bias over Variance: The "flight-by-flight variation" is often a shifting bias rather than random noise. The compromise map is specifically designed to "average out" these biases across multiple training sessions.

Key Results

  • Accuracy: The "compromise map" (using 511 inference points) produced estimates within 0.013µT of the full intermediate map (using 2001 points), proving that the computational optimization does not sacrifice precision.
  • Noise Mitigation: Tuning the controller gains and increasing sensor distance reduced the magnitude of flight-by-flight variations from roughly 1.8µT down to a range of 0.3µT – 0.5µT.
  • Consistency: The 96% consistency threshold successfully identified segments of flights where the UAV's internal bias shifted mid-air, providing a "fail-safe" for state estimation.

Limitations and Future Work

  • Time-Varying Fields: The current model assumes the indoor magnetic environment is static. It does not account for moving metallic objects like elevators or opening/closing heavy security doors.
  • Platform Specificity: The noise characteristics are specific to the ESCs and motors used. Different hardware brands might require different "quiet" gain settings.
  • Future Direction: The authors plan to integrate these maps into a real-time particle filter for 3D position localization and explore methods to update maps dynamically as the environment changes.

Summary updated

Abstract

This work presents a number of techniques to improve the ability to create magnetic field maps on a UAV which can be used to quickly and reliably gather magnetic field observations at multiple altitudes in a workspace. Unfortunately, the electronics on the UAV can introduce their own magnetic fields, distorting the resultant magnetic field map. We show methods of reducing and working with UAV-induced noise to better enable magnetic fields as a sensing modality for indoor navigation. First, some gains in our flight controller create high-frequency motor commands that introduce large noise in the measured magnetic field. Next, we implement a common noise reduction method of distancing the magnetometer from other components on our UAV. Finally, we introduce what we call a compromise GPR (Gaussian process regression) map that can be trained on multiple flight tests to learn any flight-by-flight variations between UAV observation tests. We investigate the spatial density of observations used to train a GPR map then use the compromise map to define a consistency test that can indicate whether or not the magnetometer data and corresponding GPR map are appropriate to use for state estimation. The interventions we introduce in this work facilitate indoor position localization of a UAV whose estimates we found to be quite sensitive to noise generated by the UAV.
BibTeX
@article{kuevor2023fast,
  author = {Prince Kuevor and Maani Ghaffari and Ella Atkins and James Cutler},
  title = {{Fast and Noise-Resilient Magnetic Field Mapping on a Low-Cost UAV Using Gaussian Process Regression}},
  journal = {Sensors},
  year = {2023},
  doi = {10.3390/s23083897},
}
Figure