{
  "paper_id": "78965fc35d047b2d1f966796799195b8",
  "paper_title": "FAST-LIVO: Fast and Tightly-coupled Sparse-Direct LiDAR-Inertial-Visual Odometry",
  "paper_authors": [
    "Chunran Zheng",
    "Qingyan Zhu",
    "Wei Xu",
    "Xiyuan Liu",
    "Qizhi Guo",
    "Fu Zhang"
  ],
  "publication_date": "2022",
  "publication_venue": "2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)",
  "abstract": "To achieve accurate and robust pose estimation in Simultaneous Localization and Mapping (SLAM) task, multi-sensor fusion is proven to be an effective solution and thus provides great potential in robotic applications. This paper proposes FAST-LIVO, a fast LiDAR-Inertial-Visual Odometry system, which builds on two tightly-coupled and direct odometry subsystems: a VIO subsystem and a LIO subsystem. The LIO subsystem registers raw points (instead of feature points on e.g., edges or planes) of a new scan to an incrementally-built point cloud map. The map points are additionally attached with image patches, which are then used in the VIO subsystem to align a new image by minimizing the direct photometric errors without extracting any visual features (e.g., ORB or FAST corner features). To further improve the VIO robustness and accuracy, a novel outlier rejection method is proposed to reject unstable map points that lie on edges or are occluded in the image view. Experiments on both open data sequences and our customized device data are conducted. The results show our proposed system outperforms other counterparts and can handle challenging environments at reduced computation cost. The system supports both multi-line spinning LiDARs and emerging solid-state LiDARs with completely different scanning patterns, and can run in real-time on both Intel and ARM processors. We open source our code and dataset of this work on Github to benefit the robotics community.",
  "keywords": [
    "SLAM",
    "多传感器融合",
    "激光雷达-惯性-视觉里程计（LIVO）",
    "直接法/稀疏直接图像对齐",
    "紧耦合",
    "误差状态迭代卡尔曼滤波（ESIKF）"
  ],
  "paper_institutions": [
    "The University of Hong Kong（香港大学）"
  ],
  "output_language": "zh",
  "provider": "doubao",
  "model": "glm-5.3",
  "prompt_template": "simple",
  "extracted_at": "2026-09-17T09:42:28.380362Z",
  "template_tag": "simple",
  "summary": "### 摘要\n\n针对SLAM任务中单一传感器（相机或LiDAR）在无结构、无纹理等挑战环境下难以实现精确鲁棒位姿估计的问题，以及现有LiDAR-惯性-视觉里程计系统（如R2LIVE、LVI-SAM）基于特征提取、各子系统分别处理数据而未考虑测量级耦合、计算开销大的局限，本文提出FAST-LIVO——一个快速、紧耦合、稀疏直接法的LiDAR-惯性-视觉里程计系统。该系统建立在两个直接且紧耦合的子系统之上：LIO子系统将新扫描的原始点（而非边缘或平面特征点）配准到由ikd-Tree增量组织的点云地图；VIO子系统复用LIO构建的地图点并为其附加历史观测的图像块，通过最小化直接光度误差实现帧到地图的图像对齐，无需提取ORB或FAST等视觉特征，并提出新颖的外点剔除方法以拒绝位于边缘或被遮挡的不稳定地图点。实验在NTU-VIRAL公开数据集（使用左相机、水平16通道OS1 gen1 LiDAR及其内置IMU，用于与R2LIVE、FAST-LIO2、SVO2.0和DVL-SLAM基准对比）和自采设备数据（DJI Manifold-2c机载计算机、两台MV-CA013-21UC工业相机、Livox Avia LiDAR，经STM32以10Hz触发信号硬同步，用于验证LiDAR退化、视觉挑战等极端场景）上进行。以绝对平移误差（RMSE）、端到端漂移和运行耗时为指标：FAST-LIVO在九个NTU-VIRAL序列中的八个上取得最优精度（如eee_01为0.19m、nya_03为0.11m），优于R2LIVE、FAST-LIO2、SVO2.0和DVL-SLAM；在LiDAR退化场景中端到端漂移仅0.05m，在包含剧烈运动、室内外切换和无纹理白墙的视觉挑战场景中79.52m路径的端到端误差为0.04m；完整处理平均耗时36.75ms（对比R2LIVE前端45.16ms加后端滑动窗口优化59.27ms），可在Intel与ARM处理器上实时运行——这些结果支持了直接法复用LiDAR地图点与测量级紧耦合融合的创新有效性。核心机制为误差状态迭代卡尔曼滤波（ESIKF）中的最大后验估计：min_{x_k∈M} ‖x_k ⊖ x̂_k‖²_{P̂_k} + Σ_{j=1}^{m_l} ‖r_l(x_k, ᴸp_j)‖²_{Σ_l} + Σ_{i=1}^{m_c} ‖r_c(x_k, ᴳp_i)‖²_{Σ_c}，其中LiDAR点到平面残差r_l与图像光度误差r_c分别与IMU传播紧耦合，经Gauss-Newton迭代求解。系统代码与数据集已在Github开源以支持社区进一步探索，并兼容多线旋转式与扫描模式完全不同的新兴固态LiDAR；实际部署需预先标定三传感器外参与时间偏移并保证刚性连接，在图像严重模糊（如sbs_01序列的快速UAV运动）时视觉融合对精度提升有限，且直接法依赖良好的状态初始化。\n\n### 关键要点\n\n- **方法核心**：FAST-LIVO由紧耦合的LIO与VIO两个直接法子系统构成，LIO将新扫描的原始点配准到增量构建的点云地图，VIO通过最小化直接光度误差对齐新图像，两者均不提取特征点，从而显著降低计算成本。\n- **地图点复用**：VIO子系统最大化复用LIO构建的点云地图，为地图点附加先前观测的图像块并投影到新图像进行帧到地图的稀疏直接对齐，避免了视觉特征的提取、三角化与优化，实现LiDAR与相机在测量级的耦合。\n- **外点剔除**：提出新颖的外点剔除方法，将视觉子地图点投影到当前帧并在40×40像素网格中保留最低深度点，再检查最近LiDAR扫描点在9×9邻域内的遮挡情况，剔除被遮挡或深度不连续的不稳定地图点，提升VIO鲁棒性与精度。\n- **状态估计**：采用流形上的误差状态迭代卡尔曼滤波（ESIKF），将LiDAR点到平面残差与图像光度误差分别与IMU传播在最大后验估计框架下紧耦合，通过Gauss-Newton迭代求解，等价于迭代卡尔曼滤波更新。\n- **地图管理**：LiDAR全局地图采用ikd-Tree增量组织并自动降采样与动态平衡；视觉全局地图以0.5×0.5×0.5米体素和哈希表组织，每个地图点保存位置、8×8×3像素三级图像块金字塔及对应相机位姿。\n- **基准实验设置**：定量实验在NTU-VIRAL数据集全部九个序列上进行，使用左相机、水平16通道OS1 gen1 LiDAR及其内置IMU，与R2LIVE、FAST-LIO2、SVO2.0及去除回环模块的DVL-SLAM进行公平对比。\n- **自采数据设置**：数据采集平台包含DJI Manifold-2c（Intel i7-8550u CPU、8GB RAM）、两台MV-CA013-21UC工业相机和Livox Avia LiDAR，所有传感器由STM32同步定时器以10Hz触发信号硬同步。\n- **基准结果**：在NTU-VIRAL九个序列中的八个上取得最低RMSE（0.11–0.25m），仅sbs_01因快速UAV运动导致图像严重模糊而略逊于纯LiDAR-惯性的FAST-LIO2。\n- **退化场景结果**：在面向约30米长墙面的LiDAR退化场景中，FAST-LIVO端到端漂移仅0.05m，优于在该方向缺乏约束而漂移的FAST-LIO2和SVO2.0。\n- **视觉挑战结果**：在包含室内外切换、两次剧烈运动和无纹理白墙的极难场景中，系统可靠存活并在79.52m总路径上取得0.04m的端到端误差，还可实时重建具有精细视觉细节的RGB彩色稠密地图。\n- **计算效率**：FAST-LIVO完整处理平均耗时36.75ms，远低于R2LIVE的45.16ms前端加59.27ms滑动窗口后端；VIO子系统在Intel i7上平均10.23ms、ARM上13.82ms，LIO子系统分别为26.52ms和51.51ms，具备充足的实时计算余量。\n- **适用约束**：系统假设三传感器时间偏移已知、外参预先标定且传感器刚性连接；直接法依赖良好的状态初始化；支持多线旋转式与固态LiDAR，可在Intel和ARM处理器上实时运行，代码与数据集已在Github开源。"
}