TL;DR
This paper presents a novel radio-inertial localization and tracking system for indoor environments that combines Bluetooth Low Energy (BLE) beacons with Inertial Measurement Unit (IMU) data. By developing a closed-loop system that jointly optimizes the robot's trajectory and the BLE beacon map, it achieves robust and accurate indoor positioning, even with imperfect prior maps.
What Problem Does This Paper Address?
Accurate indoor localization is crucial for many applications, but GPS signals are unavailable indoors. Traditional methods like Wi-Fi fingerprinting require extensive site surveys and can be sensitive to environmental changes. Visual-inertial systems (VINS) rely on good lighting and rich visual information, and can raise privacy concerns. While Bluetooth Low Energy (BLE) beacons offer a low-cost alternative, their signal strength is highly susceptible to multipathing and shadowing, leading to significant inaccuracies (1-10 meters). This paper aims to overcome these limitations by fusing BLE signal strength indicators with IMU data in a robust and adaptive system.
Methods
The proposed system employs a two-layer architecture: a Front-end and a Back-end.
- Front-end: This layer uses a sequential Monte Carlo (Particle Filter - PF) approach to solve the global localization and tracking problem. It leverages a novel, simplified probabilistic IMU motion model as the proposal distribution for the PF. This model predicts the robot's motion based on IMU data, making the particle sampling more efficient and accurate. BLE RSSI measurements are converted to range estimates and used to weight the particles.
- Back-end: This layer implements an incremental optimization (smoothing) algorithm (iSAM2) to solve a radio-inertial SLAM (Simultaneous Localization and Mapping) problem. It jointly estimates the robot's trajectory, the map of BLE beacon locations, and time-varying IMU bias. This back-end optimization refines the Front-end's estimates, leading to a smoother and more accurate trajectory.
The key innovation is the closed-loop architecture where the Back-end provides feedback to the Front-end. Specifically, the optimized BLE map and estimated IMU bias are fed back to the Front-end, improving its localization accuracy and robustness.
Datasets
The experiments were conducted using data collected by a wheeled mobile robot equipped with a Google Tango device (for ground truth trajectory) and a Nexus6P smartphone (for BLE, IMU, and magnetometer data). The environment was a research office space. The paper does not introduce a new benchmark dataset but uses a collected dataset for evaluation.
Key Insights and Ideas
- Closed-loop SLAM for Indoor Localization: The integration of a SLAM back-end that refines the map and sensor biases significantly improves the performance of a particle filter-based front-end. This feedback loop is crucial for robustness.
- Probabilistic IMU Motion Model: A simplified IMU motion model, used as the proposal distribution in the PF, is more efficient and accurate than random walk or dead reckoning models, especially when dealing with the inherent noise and dynamics of IMU data.
- Robustness to Map Errors: The system demonstrates remarkable robustness to errors in the prior BLE beacon map, a critical advantage for real-world deployments where perfect maps are rare.
- Fusion of Radio and Inertial Data: Combining BLE RSSI with IMU data provides a powerful approach for indoor localization, leveraging the ubiquity of BLE and the high-frequency motion information from IMUs.
Key Results
- The proposed radio-inertial system (Front-end + Back-end) significantly reduces localization error compared to PF with random walk, dead reckoning, or just the Front-end alone.
- The system achieves robust performance even when the prior BLE beacon map is perturbed by up to 3 meters.
- The closed-loop system improves localization accuracy by providing a tighter support for particle sampling and correcting IMU bias.
- Computational efficiency is demonstrated, with processing times well within the duration of the experiments, making it suitable for real-time applications.
Limitations and Future Work
- Limitations: The paper acknowledges that the system's performance can be affected by the density and distribution of BLE beacons. Non-line-of-sight (NLOS) BLE measurements are still a challenge, although the system is designed to handle noisy range measurements.
- Future Work: The authors suggest extending the system to accept any type of RSSI signals (e.g., Wi-Fi) and to explore robust observation selection techniques. A particularly promising direction is the development of a visual-radio-inertial SLAM technique, where visual data can complement radio and inertial measurements to further enhance estimation accuracy.
Summary updated