Calibrationless Radio-Inertial Localization and Tracking Systems using Smart Handheld Devices.

Maani Ghaffari, Mitesh Patel, Jaime Valls Miró, Gamini Dissanayake, Jacob T. Biehl, Andreas Girgensohn

Abstract

In this paper, we propose a novel solution for the low-cost indoor localization and tracking problem using radio signal strength indicator (RSSI), Inertial Measurement Unit (IMU), and magnetometer sensors available in smart handheld devices. The proposed algorithm does not require sensor calibration and operates in real-time. We develop a novel probabilistic IMU motion model as the proposal distribution of a sequential Monte-Carlo technique to track device trajectories. By formulating the problem as a simultaneous localization and mapping (SLAM) optimization, the system recovers continuous trajectories, precise Bluetooth Low Energy (BLE) beacon locations, and time-varying IMU bias. Experimental results demonstrate a localization error of approximately 1.35 m at the third quartile.

Introduction: The Challenge of Indoor Tracking

Indoor localization is often considered the "holy grail" for mobile robotics and smart device applications. While Visual-Inertial Navigation Systems (VINS) offer high accuracy, they are hampered by lighting conditions, high power consumption, and privacy concerns. Conversely, radio-based systems using WiFi or BLE RSSI are ubiquitous but suffer from significant noise due to signal shadowing and multi-path effects.

This research addresses three primary hurdles:

  1. Global Localization: Solving the "kidnapped robot" problem where the starting position is unknown.
  2. Sensor Noise: Mitigating the high variance of RSSI and the drifting bias of low-cost smartphone IMUs.
  3. Map Inaccuracy: Ensuring robustness when the initial positions of radio beacons are poorly surveyed or unknown.

System Architecture

The proposed system utilizes a dual-layer architecture consisting of a real-time Front-end and an optimization-based Back-end, forming a closed-loop feedback system.

The proposed closed-loop system architecture featuring a real-time Front-end and an optimization-based Back-end.

The Front-end (Tracking)

The Front-end employs a Sequential Monte-Carlo (Particle Filter) approach. A key innovation is the probabilistic IMU motion model. Instead of sampling from a complex 6D state, the filter samples tangential acceleration magnitudes along the direction of movement. This focused sampling allows the filter to maintain tracking even amidst noisy RSSI measurements.

The Back-end (Optimization)

The Back-end utilizes iSAM2 (incremental Smoothing and Mapping) to formulate the tracking problem as a factor graph. While the Front-end provides immediate estimates, the Back-end analyzes the measurement history to solve for the smoothest possible trajectory, actual BLE beacon coordinates, and internal IMU biases. These optimized values are then fed back to the Front-end to improve real-time tracking accuracy.

Experimental Evaluation

The system was tested using a wheeled mobile robot equipped with a Nexus 6P smartphone. Ground truth was established using a Google Tango device providing high-accuracy visual-inertial odometry. The testing environment was a 40m x 50m office space with active staff and standard furniture. The evaluation included 100 independent runs over trajectories exceeding 100 meters, including simulations where beacon map positions were intentionally perturbed by 3 meters to test system robustness.

Key Results

The integrated system achieved a localization error of approximately 1.35 meters at the 75th percentile (3rd quartile).

Experimental trajectory comparison on the office floor plan, showing the raw Front-end estimates versus the smoothed Back-end results against Google Tango ground truth.

Key insights from the results include:

  • Self-Calibration: By treating IMU bias and beacon locations as variables, the system eliminates the need for tedious pre-deployment surveys.
  • Trajectory Smoothing: As seen in the experimental data, the Back-end effectively corrects the "jumpy" estimates inherent in pure Particle Filters, producing a continuous path that closely tracks the ground truth.
  • Map Robustness: The system maintains high performance even with inaccurate initial maps, as the optimization process "drags" misplaced beacons toward their true physical locations based on observed motion.
  • Computational Efficiency: The MATLAB-based implementation processed 384 seconds of experimental data in under 92 seconds, confirming its suitability for real-time mobile applications.

Discussion and Future Work

While effective, the system's performance is sensitive to beacon density. In areas with sparse beacon coverage, the system relies on dead reckoning, which can lead to drift until the next beacon is encountered. Future research will focus on extending the reliable range of BLE measurements beyond the current 10-meter clamp and integrating opportunistic visual data to create a comprehensive "visual-radio-inertial SLAM" framework.

Reference: Jadidi, M. G., et al. "Calibrationless Radio-Inertial Localization and Tracking Systems using Smart Handheld Devices." arXiv:1706.05569v3 [cs.RO].

Summary updated

Abstract

In this paper, we propose a novel solution for the low-cost indoor localization and tracking problem using radio signal strength indicator, Inertial Measurement Unit (IMU), and magnetometer sensors available in smart handheld devices such as smartphones. The proposed algorithm does not require any sensor calibration and performs real-time. We develop a novel probabilistic IMU motion model as the proposal distribution of the sequential Monte-Carlo technique to track the device trajectory. Our algorithm can globally localize and track a device with a priori unknown location, given an informative prior map of the Bluetooth Low Energy (BLE) beacons. Also, we formulate the problem as an optimization problem that serves as the Back-end of the algorithm mentioned above (Front-end). Thus, by simultaneously solving for the device trajectory and the map of BLE beacons, we recover a continuous and smooth trajectory of the device, precise locations of the BLE beacons, and the time-varying IMU bias. The experimental results achieved using hardware show that through feeding back the optimized map and recovered IMU bias values to the Front-end, the overall system performance reaches to about 1.35 m at the third quartile of the cumulative distribution function of the localization error over the experimental data.
BibTeX
@inproceedings{ghaffari2018calibrationless,
  author = {Maani Ghaffari and Mitesh Patel and Jaime Valls Miró and Gamini Dissanayake and Jacob T. Biehl and Andreas Girgensohn},
  title = {{Calibrationless Radio-Inertial Localization and Tracking Systems using Smart Handheld Devices.}},
  booktitle = {Ipin},
  year = {2018},
  doi = {10.1109/ipin.2018.8533827},
}
Figure