Unmanned aerial vehicles (UAVs) are increasingly used indoors for inspection, security, and emergency tasks. Achieving accurate and robust localization under Global Navigation Satellite System (GNSS) unavailability and obstacle occlusions is therefore a critical challenge. Due to their inherent physical limitations, Inertial Measurement Unit (IMU)–based localization errors accumulate over time, Ultra-Wideband (UWB) measurements suffer from systematic biases in Non-Line-of-Sight (NLOS) environments and Visual–Inertial Odometry (VIO) depends heavily on environmental features, making it susceptible to long-term drift. We propose a tightly coupled fusion framework based on the Error-State Kalman Filter (ESKF). Using an IMU motion model for prediction, the method incorporates raw UWB ranges, VIO relative poses, and TFmini altitude in the update step. To suppress abnormal UWB measurements, a multi-epoch outlier rejection method constrained by VIO is developed, which can robustly eliminate NLOS range measurements and effectively mitigate the influence of outliers on observation updates. This framework improves both observation quality and fusion stability. We validate the proposed method on a real-world platform in an underground parking garage. Experimental results demonstrate that, in complex indoor environments, the proposed approach exhibits significant advantages over existing algorithms, achieving higher localization accuracy and robustness while effectively suppressing UWB NLOS errors as well as IMU and VIO drift.

Indoor localization for unmanned aerial vehicles (UAVs) remains challenging in GNSS-denied environments due to the difficulty of position calibration of multiple ultra-wideband (UWB) anchors and the asynchronous fusion of heterogeneous sensors. This paper proposes a multi-sensor fusion localization framework that integrates multi-anchor one-shot calibration with factor graph optimization (FGO). First, Landmark Multidimensional Scaling (LMDS) is used to reconstruct the relative geometry of the anchors and the onboard tag from ranging measurements. Then, rigid Procrustes alignment is performed using a small number of anchors with known coordinates in the East–North–Up (ENU) frame to recover the transformation to the ENU frame, thereby enabling efficient position calibration of multiple UWB anchors and UAV pose initialization. Subsequently, a tightly coupled factor graph is constructed by incorporating inertial measurement unit (IMU) pre-integration, UWB ranging, laser rangefinder height measurements, and visual–inertial odometry (VIO) pose constraints. The resulting nonlinear optimization problem is solved using incremental smoothing, which improves robustness against non-line-of-sight (NLOS) errors and long-term drift. Experimental results on anchor calibration, public datasets, and real-world indoor UAV flights demonstrate that the proposed method improves the accuracy and robustness of indoor UAV localization. In particular, on the real-world rectangle trajectory, FGO-TC reduces the RMSE by approximately 38.8% compared with FGO-LC.

Multi-robot distributed collaborative position measurement in a unified global frame has become an important research topic. Traditional multi-robot Visual-Inertial-Ranging Odometry (VIRO) utilizes inter-robot ranging to effectively correct odometry drift. However, these methods fail to consider inter-robot relative motion position relationships and weights between heterogeneous residuals, resulting in unstable position measurement accuracy along with system operation. To address this, an Error-Propagation Dilution of Precision (EPDOP)-based key trajectory point screening method is proposed, enabling efficient correction of odometry drift and ensuring global consistency without increasing the number of residual factors. In addition, a heterogeneous residual weight allocation method based on Helmert Variance Component Estimation (HVCE) is proposed, which adaptively allocates weights to improve system performance and robustness. Extensive simulations and real-world experiments demonstrate that, compared with previous methods, the proposed method reduces the average ATE by 22.21%, 28.09%, and 27.71% in dataset simulations, single-robot experiments, and multi-robot experiments, respectively, without significantly increasing the backend computation time.