Bretl Lab
2024
To address the high cost of indoor robot localization, I joined the Bretl Lab to explore low-cost methods using only smartphone-grade sensors. This project became my first step into robotics and reflected my motivation to make autonomous systems more accessible by removing the dependence on expensive sensors such as LiDAR and cameras.
Key Contributions:
- Multi-Environment Dataset: Built and collected a robust dataset across varied indoor environments, incorporating movable objects and visual-inertial SLAM for ground-truth validation
- Earth Magnetic Field Localization: Contributed to evaluation of magnetic-field “fingerprint” mapping to enable GPS-free indoor positioning using only smartphone-grade sensors.
- Low-Cost Indoor Navigation for Accessible Robotics: Demonstrated a scalable and affordable localization method, reducing reliance on expensive LiDAR/Wi-Fi systems and lowering barriers for indoor robotics
Most indoor environments—clinics, warehouses, elderly-care centers—cannot afford high-end localization systems, limiting where robots can actually be deployed. I explored an alternative approach that uses only low-cost sensors found in smartphones, mapping magnetic anomalies from building structures to enable reliable localization without installing any specialized equipment.
Motivation
GPS Positioning does not work in indoor.
Metal structures, electronic devices, and other building materials distort signals and magnetic fields, making reliable indoor localization challenging.
Alternatives are too expensive.
Systems based on LiDAR, Wi-Fi, or cameras require specialized hardware and setup, and cameras also struggle in low-light or visually degraded environments such as smoke-filled or low-visibility areas.
Magnetometers offer an attractive alternative.
They are already built into smartphones and robots, are far cheaper than LiDAR and camera systems, and operate reliably in darkness and privacy-sensitive environments.
However, indoor environments distort the Earth’s magnetic field due to steel structures, electronics, and building materials. These magnetic anomalies can instead serve as a unique “fingerprint” for reliable indoor localization.
Approach
Data was collected by navigating a mobile robot equipped with a smartphone for fiducial-marker perception and additional IMU–magnetometer sensors. This setup allowed us to record both magnetic-field anomalies and ground-truth robot trajectories across multiple indoor environments.
- Sensing Platform: Used a Lenovo Phab 2 Pro and smartphone IMU/magnetometer sensors. Inertial noise characteristics were modeled and tuned using Kalman-filter–based fusion of accelerometer and gyroscope data.
- Mapping: Collected centimeter-level ground-truth trajectories via Google Tango’s visual-inertial SLAM, combining camera features with IMU data to benchmark magnetic-field-based localization
- Calibration: Performed multi-sensor calibration across several IMU and magnetometer units to correct biases and ensure consistent reference alignment
- Robot: Used Clearpath Boxer robot and mounted the measuring component
Experiment
We conducted three types of experiments: Calibration, Dead-Load, and Live-Load
Calibration
At the North Quad of UIUC, we calibrated the sensors to estimate the local magnitude and direction of the Earth’s magnetic field and to remove magnetometer bias. The sensors—five stacked IMUs and magnetometers—were mounted on the Boxer robot. I still remember the three of us straining to lift the 300-lb Boxer robot into position—it felt like a small victory each time we managed it. That moment made the calibration process more tangible, as the data we gathered through hard- and soft-iron correction would later anchor the accuracy of every trial.
Dead-Load
The steel beams and metal structures inside buildings create small but consistent magnetic field anomalies. These anomalies become unique, location-dependent “fingerprints” that can be used for indoor localization. We placed AprilTags on walls throughout the CSL, DCL, and Talbot Lab to support visual-inertial SLAM mapping with Google Tango. Using these fiducial markers, we navigated the Boxer robot through hallways while collecting synchronized magnetometer, IMU, and ground-truth trajectory data. I’ll never forget the Talbot Lab experiments—getting last-minute approval, hauling the robot across campus at night, and racing against time to finish the runs before the building closed.
Live-Load
To simulate real-world changes, we introduced movable metal objects—such as chairs, tables, a computer cart, an industrial fan, and a toolbox—along the test path. These temporary magnetic disturbances altered the local field, allowing us to study how localization accuracy changes when the environment is no longer magnetically static. 10 scenarios of experiments were conducted to discover the disturbance of the metal objects.
Overall, the MAGPIE2 dataset generalizes well across buildings, platforms, and dynamic conditions, with recordings from three environments and both human and robot trajectories under varying live and dead loads.
Takeaways
I have always been drawn to accessible robotics, believing that one of the biggest barriers in this field is cost. In mobile robotics, localization is fundamental—but I soon learned that while GPS fails indoors, its alternatives like LiDAR and Wi-Fi are also prohibitively expensive. Through this project, I explored how to localize a robot at a fraction of the cost by leveraging the Earth’s magnetic field. What fascinated me most was turning a challenge, the magnetic distortions caused by steel structures, into a solution by using these anomalies as unique “fingerprints” for positioning. I also gained hands-on experience with SLAM using AprilTags and inertial-based localization through IMU data, deepening my understanding of practical and affordable robotic perception systems.
Further Research Idea
Expanding Application of Magnetic Localization
Magnetic-field–based localization could extend to hospitals, factories, and smart buildings to track equipment and robots using only low-cost sensors—no need for dense RF or UWB infrastructure. A promising next step is to intentionally generate artificial magnetic fields with small beacons, allowing smartphones and robots to localize even in dark, cluttered, or signal-limited environments where LiDAR or Wi-Fi fail.
Creating Intentional Magnetic Fields for Localization
A promising direction is to generate controlled magnetic fields for precise indoor positioning. This magneto-inductive approach uses small electromagnetic beacons to create predictable fields detectable by smartphones or robots, combining the stability of engineered signals with the low cost and robustness of magnetic sensing—an efficient alternative to LiDAR or RF-based localization where visual or wireless methods fall short.