Automated Calibration of RSS Fingerprinting Based Systems Using a Mobile Robot and Machine Learning

Area of Science:

• Robotics
• Machine Learning
• Indoor Positioning Systems

Background:

• Radio Signal Strength (RSS) fingerprinting is a common technique for indoor localization.
• Manual calibration of RSS fingerprinting systems is time-consuming and labor-intensive.
• Automated calibration methods are needed to improve efficiency and accuracy.

Purpose of the Study:

• To develop and evaluate an automated method for calibrating RSS-fingerprinting-based positioning systems.
• To integrate robotic platforms for data collection and machine learning for model training.
• To assess the performance of different machine learning models for radio map calibration.

Main Methods:

• Utilized a robotic platform for automated data acquisition in the system environment.
• Employed GraphSLAM for environment mapping and radio map calibration.
• Trained and compared four machine learning models: log-distance path loss, Gaussian Process Regression, Artificial Neural Network, and Random Forest Regression.
• Tested the system in a BLE-based indoor localization setup within an apartment.

Main Results:

• The automated calibration method successfully enabled positioning with accuracy comparable to existing literature.
• The Artificial Neural Network model achieved a median robot positioning error of 0.87 meters.
• The median trajectory error for a walking person localization scenario was 0.4 meters.

Conclusions:

• The proposed automated calibration method, integrating robotics and machine learning, is effective for RSS-fingerprinting positioning systems.
• The study demonstrates the feasibility of achieving high-accuracy indoor localization through automated calibration.
• The evaluated machine learning models provide viable solutions for radio map calibration and positioning.