{
  "paper_id": "33f7e51eb1f19bbe6a82bac973c5b32b",
  "paper_title": "A Robust and Fast GNSS-Inertial-LiDAR Odometry With INS-Centric Multiple Modalities by IESKF",
  "paper_authors": [
    "Jixin Gao",
    "Jianjun Sha",
    "Hongwen Li",
    "Yanheng Wang"
  ],
  "publication_date": "2024-01-09",
  "publication_venue": "IEEE Transactions on Instrumentation and Measurement",
  "abstract": "Global navigation satellite system (GNSS), inertial measurement unit (IMU), and light detection and ranging (LiDAR), as common sensors in navigation systems, are widely used in fusion simultaneous localization and mapping (SLAM) to improve the performance. However, strong robustness and high efficiency are not received much attention, posing challenges for ubiquitous application of robotics systems. This article presents a robust and fast fusion odometry with inertial navigation system (INS)-centric multiple modalities. Specifically, our framework is based on two subsystems: the GNSS-IMU subsystem estimates coarse states and provides global constraints, the LiDAR-IMU subsystem refines states and constructs the 3-D map. The state estimation is performed by iterated error state Kalman filter without interruption even if partial sensor fails. Moreover, a fast and dynamic frame transformation method and distributed backward propagation are designed for linking world-frame states with global measurements and coping with asynchronous data. Extensive qualitative and quantitative experiments are also conducted on public datasets and in real-world scenarios. The results show that our system exhibits competitive accuracy and efficiency, with strong robustness against sensor degradation and failure.",
  "keywords": [
    "Iterated error state Kalman filter (IESKF)",
    "multiple modalities",
    "multisensor simultaneous localization and mapping (SLAM)",
    "robust navigation system",
    "sensor failure"
  ],
  "paper_institutions": [
    "Qingdao Innovation and Development Base, Harbin Engineering University, Qingdao, China",
    "College of Intelligent System Science and Engineering, Harbin Engineering University, Harbin, China"
  ],
  "output_language": "zh",
  "provider": "doubao",
  "model": "glm-5.3",
  "prompt_template": "simple",
  "extracted_at": "2026-09-17T09:28:11.123990Z",
  "template_tag": "simple",
  "summary": "### 摘要\n针对多传感器融合SLAM中强鲁棒性与高计算效率难以兼顾的问题——真实环境中传感器退化或失效会使导航系统失效、基于优化器的多模态框架计算开销巨大、滤波方法又存在不可避免的误差累积——本文提出一种以惯性导航系统（INS）为核心、完全基于迭代误差状态卡尔曼滤波（IESKF）的GNSS-IMU-LiDAR融合里程计框架。系统由两个子系统构成：GNSS-IMU子系统以GNSS频率进行粗略状态估计并提供全局约束，LiDAR-IMU子系统进一步细化状态并构建三维点云地图，两子系统交替更新使状态估计在部分传感器失效时也不中断；针对坐标系不一致与多源测量异步问题，设计了基于双滑动窗口（稳定窗口与时间窗口）的快速动态坐标变换方法，在线对齐ECEF/ENU全局测量与世界系状态，并提出分布式反向传播，通过缓存GNSS预测状态对 t_{k-1} 至 t_{k+1} 间全部扫描点进行运动补偿，其核心机制为 p_{l,k+1} = ^i T_l^{-1}(^w T_i^{-1} ^{k+1}T_γ ^w T_i) ^i T_l × p_{l,γ}。实验选用含精确RTK真值与多场景的M2DGR、UrbanNav、LIO-SAM公开数据集以定量评估精度与效率，并在搭载Velodyne VLP-16 LiDAR、200 Hz JY901 IMU与4 Hz u-blox M8T GNSS的无人地面车上开展真实世界退化与失效测试，系统以C++/ROS实现并运行于Intel i7-7700HQ、8 GB内存的Ubuntu 18.04笔记本。以绝对位姿误差（APE）的RMSE与STD评估精度：11个序列中5个取得最优RMSE（如gate01为0.14 m、street03为0.13 m），其余序列仍具竞争力；以平均处理耗时评估效率：56 ms显著优于基于优化的LIO-SAM（约203–226 ms）与LINS（约112 ms），接近FAST-LIO2（约46 ms），且坐标变换在172 s序列上总耗时不足1 ms，验证了滤波式框架与快速变换方法的高效性；真实世界中LiDAR退化草地、GNSS遮挡（PDOP最大5.2）、5–10 s数据异常恢复及硬件级完全断连等场景验证了多模态设计带来的强鲁棒性。未来工作包括GNSS与LIO的更紧耦合以及多模态切换时的轨迹平滑。\n\n### 关键要点\n- 提出以INS为核心的多模态GNSS-IMU-LiDAR融合SLAM框架，完全基于IESKF实现快速状态估计，兼具竞争性精度与对传感器退化/失效的强鲁棒性。\n- 系统分为GNSS-IMU子系统（粗略估计并提供全局约束）与LiDAR-IMU子系统（细化状态并构建3D地图），两子系统交替更新保证任一传感器失效时状态估计不中断。\n- 设计基于双滑动窗口（稳定窗口 n_s≤3 与时间窗口 n_t≤5）的快速动态坐标变换方法，在线对齐GNSS全局测量与机器人局部世界系状态，172 s序列上总耗时不足1 ms。\n- 提出分布式反向传播进行运动补偿，缓存GNSS预测状态以补偿 t_{k-1} 至 t_{k+1} 间全部扫描点，避免点云碎片化与数据利用不足。\n- GNSS退化检测采用协方差阈值（c_th=45、z_th=50）及与预测状态差异阈值（2.5 m）；LiDAR退化检测以 H_l^T H_l 最小特征值为条件，退化帧被跳过。\n- GNSS失效时系统切换为LIO模态并保留ENU站心与坐标变换信息以便信号恢复后无需重新初始化；LiDAR失效时切换为GNSS-惯性里程计，地图停止更新但定位继续。\n- 实验基于M2DGR、UrbanNav、LIO-SAM公开数据集（含RTK真值）及自建UGV平台（VLP-16 LiDAR、200 Hz JY901 IMU、4 Hz u-blox M8T GNSS），在Intel i7-7700HQ、8 GB RAM、Ubuntu 18.04上以C++/ROS实现。\n- 精度评估采用APE的RMSE与STD：11个序列中5个取得最优RMSE（如gate01为0.14 m、street03为0.13 m），其余序列上FAST-LIO2与启用GPS的LIO-SAM各在3个序列更优，但本方法结果仍具竞争力。\n- 效率方面平均处理耗时56 ms，优于LIO-SAM（约203–226 ms）与LINS（约112 ms），接近FAST-LIO2（约46 ms），其中LiDAR-IMU子系统因处理大量点云占用主要计算资源。\n- 消融实验表明分布式反向传播在多数序列上提升精度，但在street04/street06因非结构化特征与动态物体导致点质量差时反而降低精度。\n- 真实世界测试覆盖LiDAR退化（开阔无结构草地）、GNSS退化（遮挡区PDOP最大5.2、SNR降至30 dB以下）、5–10 s数据异常恢复及硬件级完全断连等场景，系统均保持定位与建图功能。\n- 局限与未来方向：GNSS与LIO的更紧耦合、多模态切换时的轨迹平滑。"
}