Xinliang Li
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Xinliang Li has a total of 11 co-authors. Here are the top co-authors by number of shared publications:
- Jingbo Wei
School of Aeronautics and Astronautics, Sun Yat-sen University, Shenzhen 518107, China
3 shared publications - Kun Liu
School of Aeronautics and Astronautics, Sun Yat-sen University, Shenzhen 518107, China
3 shared publications - Zijie Qin
The School of Aeronautics and Astronautics, Sun Yat-sen University, Shenzhen, China
3 shared publications - Weijian Chen
The School of Aeronautics and Astronautics, Sun Yat-sen University, Shenzhen, China
2 shared publications - Xiaobin Yu
School of Aeronautics and Astronautics, Sun Yat-sen University, Shenzhen 518107, China
2 shared publications Chunqiang Wang
The School of Aeronautics and Astronautics, Sun Yat-sen University, Shenzhen, China
1 shared publication
Latest Publications
- 13 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 26 (7) , 10498-10512
DyGS-SLAM: Dynamic Object Segmentation and Multiview Long-Term Gaussian Consistency for High-Fidelity SLAM
Robust and accurate pose estimation in dynamic environments remains a fundamental challenge for visual simultaneous localization and mapping (SLAM). Recent advances in integrating 3D Gaussian Splatting (3DGS) into SLAM have demonstrated its potential for achieving high-quality scene reconstruction with explicit Gaussian representations. Compared with LiDAR-based systems, monocular and RGB-D cameras are lightweight, inexpensive, and capable of providing rich visual and geometric information, making them highly attractive for practical applications such as human–robot interaction and autonomous driving. However, visual SLAM systems are often sensitive to illumination variations, occlusions, and dynamic objects, which introduce unstable features that may cause pose drift and pseudo-static artifacts, thereby compromising localization stability and map integrity. To address these challenges, this paper presents DyGS-SLAM, a 3DGS-based visual SLAM framework designed for dynamic environments and compatible with both monocular and RGB-D sensors. The proposed system integrates vision-based object detection with depth sensing to perform coarse-to-fine dynamic segmentation, while a temporal-consistency refinement module leverages long-term multi-view observations to suppress residual dynamic artifacts. Extensive experiments on the TUM RGB-D and KITTI datasets show that DyGS-SLAM achieves improved localization accuracy compared with recent dynamic-aware baselines, reducing ATE by 4.7% on TUM and 6.3% on KITTI, while running in real time at 13.8 FPS with 7.4 GB GPU memory on an RTX 4080 Ti. These results highlight the potential of low-cost visual sensors for constructing high-quality 3D maps, enabling practical deployment in a wide range of intelligent sensing and robotic applications.
- 13 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 26 (7) , 10498-10512
DyGS-SLAM: Dynamic Object Segmentation and Multiview Long-Term Gaussian Consistency for High-Fidelity SLAM
Robust and accurate pose estimation in dynamic environments remains a fundamental challenge for visual simultaneous localization and mapping (SLAM). Recent advances in integrating 3D Gaussian Splatting (3DGS) into SLAM have demonstrated its potential for achieving high-quality scene reconstruction with explicit Gaussian representations. Compared with LiDAR-based systems, monocular and RGB-D cameras are lightweight, inexpensive, and capable of providing rich visual and geometric information, making them highly attractive for practical applications such as human–robot interaction and autonomous driving. However, visual SLAM systems are often sensitive to illumination variations, occlusions, and dynamic objects, which introduce unstable features that may cause pose drift and pseudo-static artifacts, thereby compromising localization stability and map integrity. To address these challenges, this paper presents DyGS-SLAM, a 3DGS-based visual SLAM framework designed for dynamic environments and compatible with both monocular and RGB-D sensors. The proposed system integrates vision-based object detection with depth sensing to perform coarse-to-fine dynamic segmentation, while a temporal-consistency refinement module leverages long-term multi-view observations to suppress residual dynamic artifacts. Extensive experiments on the TUM RGB-D and KITTI datasets show that DyGS-SLAM achieves improved localization accuracy compared with recent dynamic-aware baselines, reducing ATE by 4.7% on TUM and 6.3% on KITTI, while running in real time at 13.8 FPS with 7.4 GB GPU memory on an RTX 4080 Ti. These results highlight the potential of low-cost visual sensors for constructing high-quality 3D maps, enabling practical deployment in a wide range of intelligent sensing and robotic applications.
- 2 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 10 (6) , 5529-5536
A Quadrotor Aerial Docking System Utilizing Both Vision and Magnetic Field
This paper presents a complete quadrotor aerial docking system that utilizes both vision and magnetic field guidance to achieve high-precision docking. Visual guidance is implemented using a combination of a forward-facing camera and an upward-facing camera, which provide feedback on the local and relative positions, respectively. Magnetic guidance is realized through electromagnets on the quadrotor and permanent magnets at the docking site. The visual system provides guidance for the quadrotor to approach the target point, while the opposing magnetic poles attract and align with each other, cooperating with the quadrotor's four rotors to ensure accurate docking. In the visual guidance system, an intermediate coordinate frame is established to ensure consistent position feedback from the two sets of cameras. Meanwhile, the use of magnetic guidance enables a rapid and low-cost docking solution with greater scalability compared to mechanical docking devices. To coordinate the quadrotor's actuators with the magnetic force from the docking device, a magnetic field model is integrated into the flight controller. The model's parameters are identified offline using a simple, self-made measurement device, which collects data to train a precise model. This model is then used as a virtual sensor to provide real-time feedback. A quadratic programming (QP) problem is then formulated in the control allocation module, allowing the quadrotor's actuators to effectively coordinate with the magnetic field and guide the quadrotor to the docking point. Both simulations and experimental results validate the feasibility of the proposed system.
- 2 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 10 (6) , 5529-5536
A Quadrotor Aerial Docking System Utilizing Both Vision and Magnetic Field
This paper presents a complete quadrotor aerial docking system that utilizes both vision and magnetic field guidance to achieve high-precision docking. Visual guidance is implemented using a combination of a forward-facing camera and an upward-facing camera, which provide feedback on the local and relative positions, respectively. Magnetic guidance is realized through electromagnets on the quadrotor and permanent magnets at the docking site. The visual system provides guidance for the quadrotor to approach the target point, while the opposing magnetic poles attract and align with each other, cooperating with the quadrotor's four rotors to ensure accurate docking. In the visual guidance system, an intermediate coordinate frame is established to ensure consistent position feedback from the two sets of cameras. Meanwhile, the use of magnetic guidance enables a rapid and low-cost docking solution with greater scalability compared to mechanical docking devices. To coordinate the quadrotor's actuators with the magnetic force from the docking device, a magnetic field model is integrated into the flight controller. The model's parameters are identified offline using a simple, self-made measurement device, which collects data to train a precise model. This model is then used as a virtual sensor to provide real-time feedback. A quadratic programming (QP) problem is then formulated in the control allocation module, allowing the quadrotor's actuators to effectively coordinate with the magnetic field and guide the quadrotor to the docking point. Both simulations and experimental results validate the feasibility of the proposed system.
- 15 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 9 (12) , 11010-11017
Improving the Controller Performance of a Tilt-Rotor Octocopter by Compensating for the Tilt Angle's Dynamics
This paper presents a novel control allocation compensator for a tilt-rotor octocopter using passive hinges, aimed at enhancing attitude control performance by accounting for the dynamic characteristics of the actuators. The introduction of a passive hinging mechanism results in a slower tilt angle response compared to servo-actuated systems, which, if overlooked, can significantly degrade controller performance. In this regard, the paper first models the arm's tilting dynamics and discusses its impact on control allocation errors. A two-part compensator is then introduced to address these dynamics: the first part adds a simple control loop to the control allocation module to quicken the tilt response; the second part re-allocates the thrust based on estimated tilt angles, leveraging the motor's rapid response to offset the slow response of the tilt angle. By simultaneously accelerating the tilt response and reallocating thrusts, the compensator minimizes control allocation error, thereby enhancing the accuracy of position and attitude control. The effectiveness of the proposed method is validated through both simulations and real flight experiments.
- 15 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 9 (12) , 11010-11017
Improving the Controller Performance of a Tilt-Rotor Octocopter by Compensating for the Tilt Angle's Dynamics
This paper presents a novel control allocation compensator for a tilt-rotor octocopter using passive hinges, aimed at enhancing attitude control performance by accounting for the dynamic characteristics of the actuators. The introduction of a passive hinging mechanism results in a slower tilt angle response compared to servo-actuated systems, which, if overlooked, can significantly degrade controller performance. In this regard, the paper first models the arm's tilting dynamics and discusses its impact on control allocation errors. A two-part compensator is then introduced to address these dynamics: the first part adds a simple control loop to the control allocation module to quicken the tilt response; the second part re-allocates the thrust based on estimated tilt angles, leveraging the motor's rapid response to offset the slow response of the tilt angle. By simultaneously accelerating the tilt response and reallocating thrusts, the compensator minimizes control allocation error, thereby enhancing the accuracy of position and attitude control. The effectiveness of the proposed method is validated through both simulations and real flight experiments.