This paper proposes a novel visual-inertial-ranging odometry (VIRO) algorithm that tightly incorporates bundled ultra-wideband (UWB) sensors into existing visual-inertial odometry (VIO), with the aim to provide fully observable robot state estimation. The VIRO estimator proposed in this paper is the first VIRO estimator to demonstrate full observability.
This paper proposes a tightly coupled fusion model of visual-inertial-ranging measurements based on extended Kalman filter (EKF), where UWB anchors are integrated into a cohesive sensor network. This paper modifies the Schmidt Kalman filter formulation to efficiently incorporate the robot’s poses into state vector, facilitating online UWB network initialization and consistent state augmentation.
Observability analysis proves that the proposed VIRO system has four unobservable directions in the VIO reference frame, and achieves fully observable estimation in the gravity-aligned UWB network frame. Extensive simulation and real-world experiments are conducted to validate the proposed method.
Most existing approaches treat UWB anchors as independent sensing sources, overlooking the potential of leveraging the relative poses between these cooperative targets to enhance localization accuracy. Or, the UWB network and other VIO components are loosely coupled, with the robot pose estimation being solved independently before fusion. However, this approach fails to account for UWB sensing errors, which is unestimable, leading to manual intervention in parameter settings. This work proposed the first proved fully observable VIRO estimator based on a tightly coupled EKF framework.
