{
  "paper_id": "18df84c66d46f42d9c2990c627e9dfcf",
  "paper_title": "FAR-AVIO: Fast and Robust Schur-Complement Based Acoustic-Visual-Inertial Fusion Odometry With Sensor Calibration",
  "paper_authors": [
    "Hao Wei",
    "Peiji Wang",
    "Qianhao Wang",
    "Yang Gu",
    "Tong Qin",
    "Fei Gao",
    "Yulin Si"
  ],
  "publication_date": "",
  "publication_venue": "IEEE Robotics and Automation Letters",
  "abstract": "Underwater environments impose severe challenges to visual-inertial odometry systems, as strong light attenuation, marine snow and turbidity, together with weakly exciting motions, degrade inertial observability and cause frequent tracking failures over long-term operation. While tightly coupled acoustic-visual-inertial fusion, typically implemented through an acoustic Doppler Velocity Log (DVL) integrated with visual-inertial measurements, can provide accurate state estimation, the associated graph-based optimization is often computationally prohibitive for real-time deployment on resource-constrained platforms. Here we present FAR-AVIO, a Schur-Complement based, tightly coupled acoustic-visual-inertial odometry framework tailored for underwater robots. FAR-AVIO embeds a Schur complement formulation into an Extended Kalman Filter (EKF), enabling joint pose-landmark optimization for accuracy while maintaining constant-time updates by efficiently marginalizing landmark states. On top of this backbone, we introduce Adaptive Weight Adjustment and Reliability Evaluation (AWARE), an online sensor health module that continuously assesses the reliability of visual, inertial and DVL measurements and adaptively regulates their sigma weights, and we develop an efficient online calibration scheme that jointly estimates DVL-IMU extrinsics, without dedicated calibration manoeuvres. Numerical simulations and real-world underwater experiments consistently show that FAR-AVIO outperforms state-of-the-art underwater SLAM baselines in both localization accuracy and computational efficiency, enabling robust operation on low-power embedded platforms. Our implementation has been released as open source software.",
  "keywords": [
    "Acoustic-Visual-Inertial fusion",
    "error state Kalman filter",
    "underwater robotics"
  ],
  "paper_institutions": [],
  "output_language": "zh",
  "provider": "doubao",
  "model": "glm-5.3",
  "prompt_template": "simple",
  "extracted_at": "2026-09-17T09:37:58.154157Z",
  "template_tag": "simple",
  "summary": "### 摘要\n\n水下环境中强烈的光衰减、海洋雪与水体浑浊度以及弱激励的准静态运动，会降低惯性可观测性并导致视觉惯性里程计在长期运行中频繁跟踪失败；而现有紧耦合声学-视觉-惯性融合方法（通常将DVL与视觉惯性测量集成）虽能提供精确状态估计，但其基于图优化的后端计算开销巨大，难以在资源受限平台上实时部署，现有滤波类方法又往往将视觉与DVL信息简化为低维位姿或速度测量且不做外参在线标定。针对这一精度与效率的持续矛盾，本文提出FAR-AVIO，一个面向水下机器人的基于Schur补的紧耦合声学-视觉-惯性里程计框架，其核心机制是将Schur补消元嵌入扩展卡尔曼滤波（EKF），即通过 [b₁ − C₂C₃⁻¹b₂] = [C₁ − C₂C₃⁻¹C₂ᵀ]δX + n₁″ 将路标状态从等效观测模型中边缘化，在完整保留所有视觉测量信息的同时实现常数时间更新，兼得优化方法的精度与滤波方法的效率；在此基础上引入AWARE在线传感器健康模块，持续评估视觉、惯性与DVL测量的可靠性并自适应调节其sigma权重，同时提出无需专门标定机动的DVL-IMU外参在线联合估计方案。实验选用公开Tank数据集（八条序列，涵盖Structure/HalfTank/WholeTank三种轨迹类型与Easy/Medium/Hard三级难度，地面真值由TankGT管线基于AprilTag生成）以在真实水下条件下进行定量基准测试，并在桌面CPU（AMD Ryzen 9 7950X）与嵌入式平台（NVIDIA Jetson Orin NX）上评估计算负载，以验证低功耗部署能力。结果表明，FAR-AVIO以总RMSE 2.08 m取得所有方法中最佳平均平移精度，在困难序列上较AQUA-SLAM最多降低75%平移RMSE（如Structure-Hard上0.13 m对0.50 m），在Orin NX上单帧处理仅28.28 ms（约35 Hz），较VINS-Fusion（61.65 ms）提速约2.2倍且CPU与内存占用最低，验证了滤波后端消除优化瓶颈与AWARE自适应加权的有效性；消融实验显示启用AWARE的FAR-AVIO是唯一在所有序列上成功完成的变体，在线外参标定平均将RMSE从8.152 m降至0.263 m。未来方向包括在近静止或纯平移运动等旋转激励不足的场景下提供合理离线初值以保证标定收敛，并进一步面向AUV/ROV等水下机器人长期鲁棒运行的实际约束（无卫星导航、外部定位基础设施缺失）拓展应用。\n\n### 关键要点\n\n- FAR-AVIO是首个基于Schur补的紧耦合声学-视觉-惯性里程计框架，从多普勒频移原理严格建模DVL测量并嵌入基于滤波的后端优化，同时联合执行DVL-IMU外参在线标定，无需专门标定程序。\n- 核心机制通过Schur补消元 [b₁ − C₂C₃⁻¹b₂] = [C₁ − C₂C₃⁻¹C₂ᵀ]δX + n₁″ 将路标状态从观测模型中边缘化，在保留全部视觉测量信息的同时实现常数时间EKF更新，随后以轻量Gauss-Newton步骤更新各路标三维位置而不扩展滤波器维度。\n- AWARE模块为视觉与DVL传感器维护可靠性尺度σs与不健康事件队列，基于特征跟踪统计与重投影误差（视觉）及速度一致性与残差（DVL）计算质量分数qs∈[0,1]，动态调整融合sigma权重并在严重退化时临时禁用传感器。\n- DVL测量模型基于单束多普勒关系 vr ≈ −(cs/2ft)Δf，四波束Janus构型下通过最小二乘求解超定线性系统恢复DVL系三维速度，其协方差具有闭式解 ΣD = σ²(EᵀE)⁻¹。\n- 实验在公开Tank数据集上评估，包含八条序列、三种轨迹类型（Structure/HalfTank/WholeTank）与三级难度（由车速、光照与无纹理区域面积决定），地面真值由TankGT管线利用AprilTag标记生成。\n- 在Tank序列上，FAR-AVIO平均平移RMSE总值为2.08 m，优于AQUA-SLAM（2.42 m）、UVA-SLAM（6.41 m）、ORB-SLAM3（14.98 m）与SVIN2（94.85 m），在困难序列上较AQUA-SLAM最多降低75% RMSE。\n- 在NVIDIA Jetson Orin NX嵌入式平台上，FAR-AVIO单帧处理28.28 ms（约35 Hz），较VINS-Fusion（61.65 ms，约16 Hz）提速约2.2倍，其中视觉更新仅6.08 ms、DVL更新仅0.78 ms，CPU与内存占用在所有基线中最低。\n- 消融实验表明，启用AWARE的FAR-AVIO是唯一在所有Tank序列上成功完成的变体，而去除AWARE的版本在多数序列上跟踪失败或滤波发散；在简单序列上AWARE的尺度因子接近1，退化为标准紧耦合VI-DVL融合。\n- 在线IMU-DVL外参标定在数值仿真中从单位变换与中等/小噪声扰动初始化均收敛至真值，平均将RMSE从8.152 m降至0.263 m，在Tank数据上降低10–25%。\n- 标定收敛依赖足够的旋转激励，在近静止或纯平移运动下收敛可能缓慢，此类情况下建议提供合理的离线初始估计。\n- 纯视觉惯性方法（ORB-SLAM3、VINS-Fusion、SVIN2）在浑浊与视觉退化下频繁出现米级漂移或NaN跟踪失败，而禁用视觉的FAR-AIO在所有序列上保持功能，验证了声学-惯性融合骨干在完全视觉失效下的鲁棒性。\n- 完整实现已作为开源软件发布，适用于AUV/ROV等资源受限水下平台上海底设施检测、海洋能源维护与海洋探索等任务的实时部署。",
  "available_templates": [
    "deep_read",
    "simple"
  ]
}