UWB ranging + vision A new multi-robot co-location algorithm solution

What kind of technology is used to position bee colonies around the clock? Based on the low-cost UWB ranging and positioning technology (LinkTrack UWB high-precision positioning system) and active vision, how to realize the collaborative positioning of multiple UAVs? For the consistency problem of colocation (CL) of EKF, how to eliminate the inconsistency caused by unobservable subspace error dimensionality reduction. The research results of Dr. Hao Ning of HITCSCC team will provide a new way of thinking, the paper: KD-EKF: A Kalman Decomposition Based Extended Kalman Filter for Multi-Robot Cooperative Localization (download link at the end of the article).
In this paper, the consistency problem of EKF-based cooperative localization (CL) is studied from the perspective of Kalman decomposition, which decomposes observable and unobservable states and treats them individually. For the first time, the error variance terms leading to a reduction in the dimensionality of the unobservable subspace were explicitly separated and identified from state propagation and measurement Jacobi matrices. We show that the error difference term causes the global direction to be incorrectly observable, which in turn leads to inconsistent state estimation. A CL algorithm called EKF (KD-EKF) based on Kalman decomposition is proposed to improve consistency. The key idea is to use the Kalman observable canonical form for state estimation in the transformed coordinate system. By counteracting the error variance term, you can ensure proper observability attributes. More importantly, the modified state propagation and measurement Jacobian matrix is perfectly equivalent to linearizing a nonlinear CL system under the current best state estimation. Thus, inconsistencies caused by erroneous dimensionality reduction of unobservable subspaces are eliminated. The KD-EKF CL algorithm has been extensively validated in Monte Carlo simulations and real-world experiments, and has shown better performance than state-of-the-art algorithms in terms of accuracy and consistency.
In order to demonstrate the KD-EKF algorithm, a multi-robot system consisting of three self-developed aerial robots Hunter 2.0 is designed. Each robot is equipped with an APM flight controller, a Jetson NX embedded computer, a WIFI communication module, a Nooploop UWB module, and two fisheye cameras. The on-board UWB module and the fisheye camera (with an ultra-wide field of view of up to 200°) are used to measure the relative positions between the robots. Specifically, a YOLOv5-based target detector is implemented to provide relative azimuth observation. The detection algorithm runs at a frequency of 2Hz on the onboard computer. Combined with the ranging observation of the UWB module, the relative positions between robots can be obtained. In addition, an optimization-based visual inertial state estimator is employed to perceive the linear and angular velocities of each robot. The relative position of the robots and the self-movement information of each robot are sent to a centralized server for co-location. In addition, to evaluate the positioning results, the Norkov motion capture system is used to track the real position and attitude information of each robot.

Materials provided by the HITCSC team

Paper Link:https://arxiv.org/pdf/2210.16086.pdf