Distributed Ranging SLAM for Multiple Robots with Ultra-WideBand and Odometry Measurements

Ran Liu, Zhongyuan Deng, Zhiqiang Cao, Muhammad Shalihan, Billy Pik Lik Lau, Kaixiang Chen, Kaushik Bhowmik, Chau Yuen, U-Xuan Tan

Introduction

In many real-world scenarios, such as search-and-rescue in smoke-filled buildings or navigation through featureless tunnels, traditional Simultaneous Localization and Mapping (SLAM) sensors like LiDAR and cameras often fail. These "perceptually degraded" environments lack the visual or geometric features necessary for standard scan matching or visual odometry. Furthermore, many multi-robot systems rely on centralized architectures, which introduce significant computational bottlenecks and single points of failure.

This paper presents a distributed SLAM framework designed specifically for multi-robot teams operating in these challenging conditions. By utilizing only Ultra-WideBand (UWB) ranging and wheel odometry—sensors that are robust to lighting and environmental texture—the proposed system achieves high-accuracy localization without the need for a central server.

Methodology

The Distributed Ranging SLAM framework is composed of three integrated modules that handle the transition from raw sensor data to a consistent global map.

1. Distributed Pose Estimation

When two robots come within range of one another, they must determine their relative pose (position and orientation). UWB ranging provides distance but lacks bearing information, making the optimization problem non-convex and prone to local minima. To address this, the authors employ a coarse-to-fine approach. The system first performs a "coarse search" in polar coordinates—a method that is more efficient than a 2D grid search because UWB measurements naturally define a circular search space. Once a rough estimate is found, it is refined using non-linear least squares optimization.

2. Distributed Outlier Rejection

UWB signals are susceptible to multipath interference and Non-Line-of-Sight (NLOS) errors, which can introduce significant noise. To maintain map integrity, the framework utilizes Pairwise Consistency Maximization (PCM). This module acts as a robust filter, checking if new loop closures are mathematically consistent with the existing trajectory. By allowing robots to "vote out" erroneous measurements, the system prevents bad data from corrupting the global pose graph.

3. Distributed Pose Graph Optimization (DPGO)

To ensure scalability, the robots perform optimization locally using Riemannian gradient descent. Instead of transmitting all raw data to a central hub, robots exchange only their estimated poses when in proximity. This iterative process allows the team to reach a consensus on their collective trajectories while minimizing communication overhead.

Experimental Results

The system was validated using a team of three TurtleBot mobile robots equipped with Nooploop LinkTrack UWB nodes and wheel encoders. The experiments were conducted in a $10m \times 12m$ indoor environment, with high-precision LiDAR-based AMCL providing the ground truth.

Performance Metrics

  • Accuracy: The framework achieved a positioning accuracy of 0.45m in translation and 4.01° in rotation over trajectories of approximately 60 meters per robot.
  • Ablation Study: The coarse-to-fine strategy proved essential; using standard non-linear optimization alone resulted in translation errors of ~8m, whereas the proposed method reduced initial estimation errors to ~4.5m before final graph optimization.
  • Efficiency: The system is highly optimized for real-time use, with a total computational time of 81ms across all modules.
  • Communication: The distributed nature of the algorithm kept data exchange low, totaling only 13.8 MB for the entire mission at a 10Hz update rate.

Conclusion

This research demonstrates that robust multi-robot SLAM is achievable using "thin" sensor data like UWB ranging and odometry. By combining polar-based pose searching with robust outlier rejection and distributed optimization, the system provides a scalable solution for environments where traditional sensors fail.

Future work will focus on extending the framework from 2D to 3D environments to support aerial platforms and exploring more complex quadratically constrained quadratic programming (QCQP) solutions to further improve initial pose estimation.

Summary updated

BibTeX
@inproceedings{liu2022distributed,
  author = {Ran Liu and Zhongyuan Deng and Zhiqiang Cao and Muhammad Shalihan and Billy Pik Lik Lau and Kaixiang Chen and Kaushik Bhowmik and Chau Yuen and U-Xuan Tan},
  title = {{Distributed Ranging SLAM for Multiple Robots with Ultra-WideBand and Odometry Measurements}},
  booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems},
  year = {2022},
  doi = {10.1109/iros47612.2022.9982028},
}
Figure