Multi-Robot Relative Pose Estimation in SE(2) With Observability Analysis: A Comparison of Extended Kalman Filtering and Robust Pose Graph Optimization

Abstract

In this study, we focus on multi-robot cooperative localization and the observability analysis of relative pose estimation. Cooperative localization enhances pose estimation by sharing odometry data via communication networks. When a target robot's odometry is directly shared, relative pose observability can be achieved using either range-only or bearing-only measurements, given non-zero linear velocities. However, if the target robot's odometry must be estimated, both range and bearing measurements are required for observability. We validate relative pose estimation feasibility in ground-based multi-robot systems through simulations and real-world experiments, exploring six different sensing and communication structures. In ROS/Gazebo simulations, we evaluate estimation accuracy using extended Kalman filter (EKF) and pose graph optimization (PGO) methods with various robust loss functions and sliding window batch sizes. Hardware experiments with two Turtlebot3 robots equipped with UWB modules further demonstrate the practicality of decentralized relative pose estimation without inter-robot communication. The results show that estimation accuracy decreases by only 0 to 8% compared to setups utilizing inter-robot communication, underscoring the effectiveness of our proposed methods.