D-GVIO: A Buffer-Driven and Efficient Decentralized GNSS-Visual-Inertial State Estimator for Multi-Agent Systems

Yarong Luo; Wentao Lu; Chi Guo; Ming Li

cs.ROMar 2, 2026

D-GVIO: A Buffer-Driven and Efficient Decentralized GNSS-Visual-Inertial State Estimator for Multi-Agent Systems

Authors: Yarong Luo, Wentao Lu, Chi Guo, Ming Li

Organizations: School of Robotics, Wuhan University, Wuhan 430072, China. · School of Electronic Information of Wuhan University, Hubei Luojia Laboratory, Wuhan 430072, China.

Abstract

Cooperative localization is essential for swarm applications like collaborative exploration and search-and-rescue missions. However, maintaining real-time capability, robustness, and computational efficiency on resource-constrained platforms presents significant challenges. To address these challenges, we propose D-GVIO, a buffer-driven and fully decentralized GNSS-Visual-Inertial Odometry (GVIO) framework that leverages a novel buffering strategy to support efficient and robust distributed state estimation. The proposed framework is characterized by four core mechanisms. Firstly, through covariance segmentation, covariance intersection and buffering strategy, we modularize propagation and update steps in distributed state estimation, significantly reducing computational and communication burdens. Secondly, the left-invariant extended Kalman filter (L-IEKF) is adopted for information fusion, which exhibits superior state estimation performance over the traditional extended Kalman filter (EKF) since its state transition matrix is independent of the system state. Thirdly, a buffer-based re-propagation strategy is employed to handle delayed measurements efficiently and accurately by leveraging the L-IEKF, eliminating the need for costly re-computation. Finally, an adaptive buffer-driven outlier detection method is proposed to dynamically cull GNSS outliers, enhancing robustness in GNSS-challenged environments.

Figures & tables

Fig. 1: Decentralized GVIO system with two agents (

\textrm{Agent}_{1}

\textrm{Agent}_{2}

). Each agent runs a local filter for propagation and private updates using IMU, GNSS, and camera data. Collaborative updates occur when both agents observe common landmarks via feature matching.

Fig. 2: Block diagram of the proposed system. Each agent maintains local SLAM features and MSCKF tracks for measurement updates. When receiving messages from other agents, the system finds feature correspondences and performs collaborative MSCKF or SLAM updates.

Fig. 3: Coordinate frames in the GVIO system and their transformations between each other. The transformation of frame

\{C\}

into the frame

\{I\}

\left(\bm{R}^{I}_{c},\bm{p}^{I}_{c}\right)

and the transformation of frame

\{I\}

into the frame

\{W\}

\left(\bm{R}^{w}_{I},\bm{p}^{w}_{I}\right)

Fig. 4: Cross-covariance propagation is delayed until the next update step (at time

t_{k}

), and is propagated using the state transition matrix and correction term stored in the buffer.

Fig. 5: Delayed measurement scenario: Sensor 1 performs update step at time

t_{k}

. Due to the communication and computational delays, the update step is not completed until time

t_{m}

Algorithm 1: Position Constraint Outlier Culling
Input: GNSS positions $\bm{p}^{GNSS}_{k}$ , VIO positions $\bm{p}^{VIO}_{k}$ , $\mathcal{B}^{GNSS}$
GNSS buffer size $W$ , Parameters $\beta$ and $\bm{v}^{emp}$
1. Preprocessing:
Compute GNSS velocity squared:
${\bm{v}^{GNSS}_{k}}=\frac{1}{\Delta t}\sum_{k=1}^{3}(\bm{p}^{GNSS}_{k}-\bm{p}^{GNSS}_{k-1})$
Compute VIO velocity squared:

Table 6

Fig. 6: Dataset scenarios (a) "Castle Around" four drones fly around the Inveraray Castle model (https://skfb.ly/6z7Rr) by Andrea Spognetta licensed under Creative Commons Attribution-NonCommercial. (b) "S3E Square" two wheeled robots driving around the square. (c) "our real-world dataset" two wheeled robots driving in parallel around a building. The subplot within the yellow box highlights GNSS outliers.

Datasets	Agent	Ours	X-VIO	COVINS	Ours	InGVIO	IC-GVINS
		(VIO)	(VIO)	(VIO)	(GVIO)	(GVIO)	(GVIO)
Castle Around	UAV 1	$0.67$	$0.70$	$0.51$	0.07	$0.08$	0.06
	UAV 2	$0.74$	$0.73$	$0.63$	$0.11$	0.07	0.06
	UAV 3	$0.82$	$0.82$	$0.76$	0.08	$0.13$	0.11
	UAV 4	$0.79$	$0.83$	$0.80$	0.07	$0.09$	0.10

TABLE I: ATE (meter) comparison with the SOTA on the Castle Around dataset.

Algorithm	Agent	N. of Messages sent	Max CPU (%)	Max Mem (MiB)
Ours	UAV 1	1042	135.66	81.21
	UAV 2	1068	142.01	89.31
	UAV 3	1231	$167.75$	95.22
	UAV 4	1256	171.68	92.31
COVINS	UAV 1	$3568$	$200.10$	657.89
	UAV 2	$3576$	$205.21$	648.48

TABLE II: Algorithm performance comparison with the SOTA on the Castle Around dataset.

Fig. 7: The ANEES of estimated position and orientation using different filters.

Fig. 8: The ATE of D-GVIO using different filters on the S3E Square dataset.

Fig. 9: The execution time of D-GVIO and centralized EKF. The legends inform about the average execution time for each robot. The two plots in the first row and second rows show the execution time of propagation and vision-related private and collaborative measurement update steps for two robots, respectively.

Fig. 10: Trajectories estimate results for Robot 3 and Robot 4 in the real-world dataset.

Robot State Estimation

Robot state estimation infers a robot’s position, motion, orientation, and internal configuration from noisy sensors and physical constraints.

See all 200 papers

Distributed State Estimation for Vision-Based Cooperative Slung Load Transportation in GPS-Denied Environments
Mar 4, 2026Jack R. Pence, Jackson Fezell, Jack W. Langelaan +1Aerial Robotics Relative Pose Estimation
The Pennsylvania State University, University Park, Pennsylvania, 16802
Distributed and Consistent Multi-Robot Visual-Inertial-Ranging Odometry on Lie Groups
Feb 22, 2026Ziwei Kang, Yizhi ZhouRobot State Estimation Robot Localization
School of Control and Computer Engineering, North China Electric Power University, Beijing, VA 102206, China · Department of Electrical and Computer Engineering, George Mason University, Fairfax, VA 22030, USA
Decentralized Cooperative Localization for Multi-Robot Systems with Asynchronous Sensor Fusion
Mar 12, 2026Nivand Khosravi, Niusha Khosravi, Mohammad Bozorg +1Cooperative Localization
Department of Electrical and Computer Engineering, Instituto Superior T´ecnico, Lisbon, Portugal · Department of Mechanical Engineering, Yazd University, Yazd, Iran · Department of Mechanical Engineering, University of Birmingham, Birmingham, United Kingdom

Mar 4, 2026J/K move · Enter open · S save

Distributed State Estimation for Vision-Based Cooperative Slung Load Transportation in GPS-Denied Environments

Jack R. Pence, Jackson Fezell, Jack W. Langelaan, Junyi Geng

The Pennsylvania State University, University Park, Pennsylvania, 16802

Aerial Robotics Relative Pose Estimation Multi-Agent Coordination

Open paper AskPDF arXiv

Transporting heavy or oversized slung loads using rotorcraft has traditionally relied on single-aircraft systems, which limits both payload capacity and control authority. Cooperative multilift using teams of rotorcraft offers a scalable and efficient alternative, especially for infrequent but challenging "long-tail" payloads without the need of building larger and larger rotorcraft. Most prior multilift research assumes GPS availability, uses centralized estimation architectures, or relies on controlled laboratory motion-capture setups. As a result, these methods lack robustness to sensor loss and are not viable in GPS-denied or operationally constrained environments. This paper addresses this limitation by presenting a distributed and decentralized payload state estimation framework for vision-based multilift operations. Using onboard monocular cameras, each UAV detects a fiducial marker on the payload and estimates its relative pose. These measurements are fused via a Distributed and Decentralized Extended Information Filter (DDEIF), enabling robust and scalable estimation that is resilient to individual sensor dropouts. This payload state estimate is then used for closed-loop trajectory tracking control. Monte Carlo simulation results in Gazebo show the effectiveness of the proposed approach, including the effect of communication loss during flight.

Abstract

Robot State Estimation

Explore similar work

Distributed State Estimation for Vision-Based Cooperative Slung Load Transportation in GPS-Denied Environments

Distributed and Consistent Multi-Robot Visual-Inertial-Ranging Odometry on Lie Groups

Decentralized Cooperative Localization for Multi-Robot Systems with Asynchronous Sensor Fusion