13 May 2024 in IEEE Robotics and Automation Letters
Visual simultaneous localization and mapping (VSLAM) has broad applications, with state-of-the-art methods leveraging deep neural networks for better robustness and applicability. However, there is a lack of research in fusing these learning-based methods with multi-sensor information, which could be indispensable to push related applications to large-scale and complex scenarios. In this letter, we tightly integrate the trainable deep dense bundle adjustment (DBA) with multi-sensor information through a factor graph. In the framework, recurrent optical flow and DBA are performed among sequential images. The Hessian information derived from DBA is fed into a generic factor graph for multi-sensor fusion, which employs a sliding window and supports probabilistic marginalization. A pipeline for visual-inertial integration is firstly developed, which provides the minimum ability of metric-scale localization and mapping. Furthermore, other sensors (e.g., global navigation satellite system) are integrated for driftless and geo-referencing functionality. Extensive tests are conducted on both public datasets and self-collected datasets. The results validate the superior localization performance of our approach, which enables real-time dense mapping in large-scale environments.
25 January 2024 in IEEE Transactions on Vehicular Technology
Multi-agent cooperative positioning (CP) is widely applied in many applications of unmanned ground vehicles and unmanned aerial vehicles. Global Navigation Satellite System (GNSS) is an effective tool to conduct CP on a large scale. However, the performance of traditional GNSS-based CP algorithms is limited by the accuracy of single-agent self-positioning. In this paper, we propose a robust GNSS-based cooperative algorithm to achieve instantaneous precise positioning of multiple agents. In the proposed algorithm, precise point positioning-real-time kinematic (PPP-RTK) is utilized to conduct single-agent self-positioning, which is capable of achieving centimeter-level positioning rapidly with precise atmospheric corrections. Moreover, inter-agent associations derived from double-differenced carrier-phase observations and inter-epoch associations obtained via time-differenced carrier-phase observations are integrated with single-agent self-positioning by a factor graph-based optimizer. To evaluate the proposed scheme, real-world experiments with four agents in different scenarios were carried out. Based on the 30-minute field experimental results, the proposed scheme showed superior positioning performance with minimum and maximum three-dimension root mean square errors (3D-RMSEs) of 5.9 cm and 10.3 cm, respectively. It turned out that benefiting from PPP-RTK's characteristic of achieving precise positioning, the proposed method can converge to centimeter-level accuracy instantaneously.
17 September 2025 in IEEE Internet of Things Journal
Precise position information plays a crucial role in industries such as intelligent autonomous systems and autonomous driving. Combining the advantages of traditional precise point positioning (PPP) and real-time kinematic (RTK) technologies, the emerging PPP-RTK technique is able to provide real-time centimeter-level positioning in open environments. However, its stability is often insufficient in complex environments due to frequent interference of satellite signals. To enhance the continuity and robustness of positioning results in such challenging environments, an inertial navigation system (INS), which can perceive the motion information of the carrier, is commonly integrated with a global navigation satellite system (GNSS). In this article, an optimization-based tightly coupled multi-GNSS PPP-RTK/INS system is proposed to achieve continuous and reliable positioning in urban environments. In the proposed algorithm, an estimator based on maximum a posteriori (MAP) estimation is utilized to integrate multi-GNSS raw observations, inertial measurements, and precise atmospheric information. Moreover, the outlier detection algorithm based on standardized residuals is applied in the proposed method to reject gross errors introduced by GNSS observations, making the positioning results more consistent with observations to get reliable ambiguity-float solutions. Then, a cascaded ambiguity resolution is conducted to obtain final, precise ambiguity-fixed solutions. The experimental results show that the proposed method achieves centimeter-level horizontal position estimates in urban environments, with an accuracy improvement of over 80% compared to PPP-RTK. Besides, the proposed optimization-based algorithm represents an accuracy improvement of approximately 30% compared to the traditional extended Kalman filter (EKF)-based method.
17 March 2026 in IEEE Transactions on Aerospace and Electronic Systems
Ionospheric delay, as a significant error source in GNSS raw observations, deserves rigorous handling in single-frequency precise point positioning (SF-PPP). Currently, SF-PPP commonly employs global ionospheric maps (GIM) from IGS analysis centers, obtaining slant delays through spatiotemporal interpolation and oblique projection. However, constrained by inaccuracies in phase-smoothed pseudorange-derived slant total electron content (STEC) and truncation errors during fitting procedures, GIM products achieve only 0.32-1.28 m equivalent ranging accuracy at GPS L1 frequency, offering limited enhancement for SF-PPP. In this paper, we propose a lightweight multilayer perceptron (MLP)-based ionospheric modeling approach to enhance SF-PPP. The proposed method utilizes undifferenced and uncombined PPP to generate raw slant ionospheric delays as training labels. The input features are constructed from satellite-related parameters (e.g. time, locations of the ionosphere pierce points, elevation and azimuth angles) that are readily available during precise positioning computation. By leveraging the exceptional nonlinear fitting capability of MLP, the proposed model can effectively capture the relationship between the slant ionospheric delay and satellite-related parameters to enhance SF-PPP. Experimental results demonstrate that the MLP-based approach achieves slant delay mean absolute error below 0.3 m. Compared to final ionospheric products from Center for Orbit Determination in Europe (CODE), the proposed method demonstrates a 14.9% to 52.6% reduction in slant ionospheric delay root mean square error across different time periods. Furthermore, the proposed method significantly reduces convergence time while enhancing positioning accuracy, with 95th percentile horizontal and vertical positioning errors below 0.527 m and 0.831 m respectively - corresponding to 41.5% and 43.3% reductions compared to CODE GIM-enhanced SF-PPP.
29 August 2025 in IEEE Sensors Journal
Online camera calibration is essential for maintaining navigation stability by compensating for drift or unknown calibration parameters in Visual-Inertial Navigation Systems (VINS). However, it remains challenging due to the limited availability of valid information during motion. In this paper, we present a novel online camera calibration method for VINS by tightly coupling Global Navigation Satellite System (GNSS) information. Specifically, the proposed approach enables stable and accurate online calibration of both camera intrinsic and camera-IMU extrinsic parameters by incorporating GNSS pseudorange and carrier-phase measurement constraints into a centralized Extended Kalman Filter (EKF). The multi-sensor fusion mechanism enhances state estimation consistency through sensor complementarity, enabling robust navigation performance alongside dynamic parameter calibration. The results from both simulations and real-world experiments demonstrate that GNSS assistance enhances state observability, accelerates convergence, and improves the correction of calibration errors at various scales. In a 230-meter outdoor navigation task, the calibration results from different methods were treated as fixed parameters for evaluation. The proposed method achieved an altitude drift of 1.95 m and an absolute pose error (APE) root mean square error (RMSE) of 3.7 m, outperforming both Kalibr offline calibration (2.75 m drift, 3.98 m RMSE) and GNSS-absent online calibration (7.98 m drift, 4.98 m RMSE).
10 December 2025 in IEEE/ASME Transactions on Mechatronics
In the domain of autonomous systems, the integration of multiple heterogeneous sensors represented by global navigation satellite system (GNSS), cameras, and inertial measurement units (IMU) has been widely employed to achieve robust navigation performance. In addition, with the miniaturization and cost reduction of sensors, homogeneous sensors' integration platforms become available for both manufacturing and application. This article proposes a tightly coupled system with multiple GNSS terminals and multiple IMUs, aiming to improve vehicle navigation performance. First, a weighted least squares based synthetic IMU generator is designed to fuse multiple IMU measurements to attenuate the noise. Besides, double-difference pseudorange and carrier phase measurements from multiple GNSS, along with measurements from the low-noise synthetic IMU, are tightly fused in a stacked estimator. This enriches GNSS observations with enhanced satellite geometry and improves IMU propagation accuracy. Through real-world onboard experiments and a simulation test in different scenarios, it is shown that in GNSS-challenged environments, the proposed method maintains horizontal centimeter-level positioning accuracy and achieves an ambiguity resolution rate exceeding 90%. Even in GNSS-denied environments, it can significantly suppress the drift of positioning errors.