Multi-Robot Relative Pose Estimation and IMU Preintegration Using Passive UWB Transceivers

TL;DR

The paper tackles scalable, high-accuracy relative pose estimation for multi-robot teams using UWB ranging and IMU data. It introduces a passive-listening ranging protocol and an on-manifold EKF that fuses ToF/ranging measurements with preintegrated IMU data directly on SE_2(3), enabling clock synchronization and relative pose estimation across robots. Through simulations and experiments with up to 3 quadcopters and multiple trials, the approach achieves substantial localization improvements (up to 48% over no passive listening and approaching centralized performance) while maintaining robustness to communication constraints. The proposed methods offer practical benefits for multi-robot autonomy, including higher measurement throughput, simpler MAC design, and efficient sharing of odometry via RMI preintegration, with extensions to incomplete graphs and IMU biases discussed for real-world deployment.

Abstract

Ultra-wideband (UWB) systems are becoming increasingly popular as a means of inter-robot ranging and communication. A major constraint associated with UWB is that only one pair of UWB transceivers can range at a time to avoid interference, hence hindering the scalability of UWB-based localization. In this paper, a ranging protocol is proposed that allows all robots to passively listen on neighbouring communicating robots without any hierarchical restrictions on the role of the robots. This is utilized to allow each robot to obtain more range measurements and to broadcast preintegrated inertial measurement unit (IMU) measurements for relative extended pose state estimation directly on SE2(3). Consequently, a simultaneous clock-synchronization and relative-pose estimator (CSRPE) is formulated using an on-manifold extended Kalman filter (EKF) and is evaluated in simulation using Monte-Carlo runs for up to 7 robots. The ranging protocol is implemented in C on custom-made UWB boards fitted to 3 quadcopters, and the proposed filter is evaluated over multiple experimental trials, yielding up to 48% improvement in localization accuracy.

Figures (15)

• Figure 1: The experimental set-up.
• Figure 2: The trajectories followed by three simulated quadcopters.
• Figure 3: The distribution of the posterior position of the green robot given a position prior and a single range measurement with the red robot.
• Figure 4: A summary of the operators between elements of the different spaces associated with $SE_2(3)$.
• Figure 5: An example of a ranging transaction, where Transceivers $f_1$ and $s_2$ are actively ranging with one another and all other tags are passively listening.