Hector SLAM
Hector SLAM 結合以 LiDAR 為主的 2D 掃描對地圖（scan-to-map）匹配與以 IMU 為主的 3D 姿態估計：先用估計姿態把掃描轉到穩定座標系，再以 Gauss-Newton 在雙線性內插的佔據網格上求位姿，並用多解析度網格降低陷入局部極小的風險。作者表示在所考慮的小尺度情境中精度足以不需明確的迴圈閉合，並以手持系統在救援競技場與新建築中建圖。
本頁內容
LiDAR scan-to-grid Gauss-Newton matching with multi-resolution maps plus an IMU-based 6-DoF EKF, running without odometry or explicit loop closure in small-scale scenes.
技術屬性
欄位內容為文獻擷取紀錄的原文用語（英文），以原文為據；「未查證」表示本研究尚未讀到該資訊，不代表該方法不具備此能力。
| 感測輸入 | 2D laser range finder、IMU |
|---|---|
| 原文測試平台 | wheeled UGV、handheld、USV (surface vessel) |
| 狀態估計 | Gauss-Newton scan-to-map matching on bilinearly interpolated multi-resolution occupancy grids (coarse to fine), with match covariance from the approximate Hessian; 6-DoF EKF navigation filter at 100 Hz with gyro and accelerometer bias states, fused with the SLAM pose by covariance intersection; the EKF pose projected to the plane serves as the scan-matcher start estimate (Sec. IV-B, IV-C, V) |
| 資料關聯 | scan endpoints matched to bilinearly interpolated occupancy grid gradients (no feature extraction) |
| 時間表示 | discrete poses |
| 去畸變 | 原文未報告 (scans are transformed to a stabilized frame using estimated attitude; intra-scan motion compensation not described) |
| 迴圈閉合 | none (authors report that explicit loop closing was not required in the considered small-scale scenarios) |
| 全域最佳化 | none |
| 地圖表示 | multi-resolution 2D occupancy grids, used optionally, each coarser level at half the resolution and all levels updated simultaneously from the estimated poses, with matching started at the coarsest level (examples: 20, 10 and 5 cm cells in Fig. 3; 0.25 m grid in the USV map, Fig. 5b) (Sec. IV-C) |
| 先驗資訊 | none |
| 可輸出幾何 | 2D occupancy grid map and 2D pose; 6-DoF attitude from EKF |
| 計算需求 | logged data processed at 3x real time on an Intel Atom Z530 embedded board, using less than half of that system's resources (Sec. VI.C) |
使用設備
原文使用的感測器、運算硬體與載具（equipment）。型號保留原文寫法，連結到設備頁中同一型號的歸併名稱；角色依原文用途分為方法輸入、資料集感測器、執行運算平台、參考或真值量測（reference or ground truth）與比較對象設備。
| 類別 | 型號（原文寫法） | 角色 | 資料集 | 原文規格 | 出處 |
|---|---|---|---|---|---|
| LiDAR | Hokuyo UTM-30LX | 方法輸入 | 未標示 | returns no valid distance when beams hit water | (Kohlbrecher et al., 2011, Sec. VI-B, VI-C) |
| 行動掃描設備 | self-contained embedded mapping system for handheld mapping | 方法輸入 | 未標示 | Hokuyo UTM-30LX, Intel Atom Z530 board and MEMS IMU; can be carried by hand or mounted on vehicles | (Kohlbrecher et al., 2011, Sec. VI-C, Fig. 4b) |
| 慣性量測單元（IMU） | small low-cost MEMS IMU (model not reported) | 方法輸入 | 未標示 | 原文未報告 | (Kohlbrecher et al., 2011, Sec. VI-C) |
| 載具平台 | Hector UGV | 方法輸入 | 未標示 | LIDAR stabilized about roll and pitch to stay aligned with the ground plane | (Kohlbrecher et al., 2011, Sec. VI-A, Fig. 4a) |
| 載具平台 | USV platform (model not reported) | 方法輸入 | 未標示 | carried the Hector UGV mapping system on Claytor Lake, Virginia | (Kohlbrecher et al., 2011, Sec. VI-B, Fig. 4c) |
| 運算硬體 | Intel Atom Z530 based CPU board | 執行運算平台 | 未標示 | embedded board of the handheld mapping system | (Kohlbrecher et al., 2011, Sec. VI-C) |
作者報告的優勢與限制
優勢
- Low computational cost suitable for embedded, handheld and aerial use: logged data replayed at 3x real time on an Intel Atom Z530 using under half of its resources (abstract
- Sec. VI-C) | Does not require wheel odometry or extracted features
- worked in unstructured coastal vegetation (Sec. VI-B) | Mounted on a USV without modification and operational in under 30 minutes
- the Hokuyo UTM-30LX returns no valid ranges on water, which kept water returns out of SLAM (Sec. VI-B)
限制
- Intended for small-scale scenarios where large loops do not have to be closed (Sec. I) | Gradient-based matching is prone to local minima, mitigated but not eliminated by multi-resolution maps (Sec. IV-C) | Gauss-Newton works on non-smooth linear approximations of the map gradient, so local quadratic convergence is not guaranteed (Sec. IV-B) | Full 3D position needs an additional height sensor such as a barometer or range sensor (Sec. V-A) | USV maps were built without GPS
- pose-graph optimization with GPS constraints is left for future work (Sec. VI-B) | Authors note standard benchmarks lack the 6-DoF motion and high LiDAR rate they target, so they provide their own bag files instead (Sec. VI)
營建工程相關證據
未在施工中工地測試。手持系統曾在 RoboCup 2011 救援競技場與 Dagstuhl 城堡的新建築中建圖，並與參考地圖疊合比對（Fig. 6）；致謝指出 Dagstuhl 的參考資料是由他人提供的樓層平面圖，競技場參考地圖來源未說明。文中只有定性疊合，沒有量化誤差，不能作為工程幾何精度證據。
原文驗證環境：受控實驗、已完工建築、跨場域
報告的性能數據
以下是原文作者報告的性能數值（author-reported results），不是本研究重新量測的結果。每張圖只並列同一個比較組（comparison group，同一張表、同一組實驗設定）內的方法；不同比較組之間的數值不可直接比較，也不構成排名。
本方法共出現在 4 個比較組，合計 12 筆紀錄。
Labbé & Michaud, 2019 · Table 9 本方法 8 筆
表格設定（擷取紀錄原文）：MIT Stata Center 2012-01-25 sequences; RTAB-Map WheelIMU→S2M versus other ROS 2D lidar SLAM run with default parameters; Cartographer, GMapping and Karto use WheelIMU odometry, Hector SLAM uses none; GMapping ATE computed on the current best particle path (Labbé & Michaud, 2019, Table 9)
ATEend，MIT Stata Center (PR2) · 2012-01-25-12-14-25
只並列這張表在相同設定下報告的方法；以「本方法：」開頭者為本頁方法。失敗、未執行與未報告以標記呈現，不是 0。
按 Tab 進入圖表後，用上下方向鍵逐一瀏覽各類別，Esc 關閉提示框；也可開啟表格檢視閱讀全部數值。
這些是 Labbé & Michaud, 2019 在此表設定下報告的數值（author-reported results），只能在同一個比較組內對照，不代表方法在其他資料或設定下的表現。
資料來源作者報告值(Labbé & Michaud, 2019, Table 9)
| 方法（原文寫法） | 報告值 | 出處 |
|---|---|---|
| RTAB-Map (WheelIMU→S2M) [Long-range lidar]原文提出 | 0.05 m | (Labbé & Michaud, 2019, Table 9) |
| Cartographer (WheelIMU) [Long-range lidar] | 0.11 m | (Labbé & Michaud, 2019, Table 9) |
| GMapping (WheelIMU) [Long-range lidar] | 0.19 m | (Labbé & Michaud, 2019, Table 9) |
| Karto SLAM (WheelIMU) [Long-range lidar] | 0.22 m | (Labbé & Michaud, 2019, Table 9) |
| Hector SLAM (no odometry) [Long-range lidar]本方法 | 0.06 m | (Labbé & Michaud, 2019, Table 9) |
| RTAB-Map (WheelIMU→S2M) [Short-range lidar]原文提出 | 0.07 m | (Labbé & Michaud, 2019, Table 9) |
| Cartographer (WheelIMU) [Short-range lidar] | 0.45 m | (Labbé & Michaud, 2019, Table 9) |
| GMapping (WheelIMU) [Short-range lidar] | 1.71 m | (Labbé & Michaud, 2019, Table 9) |
| Karto SLAM (WheelIMU) [Short-range lidar] | 0.48 m | (Labbé & Michaud, 2019, Table 9) |
| Hector SLAM (no odometry) [Short-range lidar]本方法 | 4.59 m | (Labbé & Michaud, 2019, Table 9) |
Kohlbrecher et al., 2011 · Text Sec. VI-C 本方法 2 筆
資料集與序列RoboCup 2011 Rescue Arena and Dagstuhl new building logs
表格設定（擷取紀錄原文）：Handheld embedded mapping system; logged sensor data replayed to the SLAM system on the Atom Z530 CPU (Kohlbrecher et al., 2011, Text Sec. VI-C)
playback speed without loss of map quality，RoboCup 2011 Rescue Arena and Dagstuhl new building logs
這張表在此指標與資料序列只列出本方法一筆，沒有可並列的其他方法，因此不畫圖，數值與出處見下表。這是 Kohlbrecher et al., 2011 在此表設定下報告的數值（author-reported results），不代表方法在其他資料或設定下的表現。
| 方法（原文寫法） | 報告值 | 出處 |
|---|---|---|
| Hector SLAM本方法原文提出硬體：Intel Atom Z530 based CPU board | 3 x real time | (Kohlbrecher et al., 2011, Sec. VI-C) |
Chen et al., 2025a · Table 7 本方法 1 筆
指標Position Drift (m)
資料集與序列authors' HKUST corridor recording (ROS bag) · single inspection trip, 92.77 m, 195.37 s
表格設定（擷取紀錄原文）：Localization RMSE computed with EVO against a reference built from loop closure plus scale alignment to the TLS cloud; one 92.77 m quadruped trip (Go1 Edu) in an academic-building corridor; alignment mode not stated (Chen et al., 2025a, Table 7)
Position Drift (m)，authors' HKUST corridor recording (ROS bag) · single inspection trip, 92.77 m, 195.37 s
這張表在此指標與資料序列只列出本方法一筆，沒有可並列的其他方法，因此不畫圖，數值與出處見下表。這是 Chen et al., 2025a 在此表設定下報告的數值（author-reported results），不代表方法在其他資料或設定下的表現。
| 方法（原文寫法） | 報告值 | 出處 |
|---|---|---|
| Hector SLAM [Laser]本方法 | 無數值未報告註記（擷取紀錄）：not reported ('-' in all Table 7 cells); Sec. 4.2.2 says Hector SLAM could lose track during direction change and is subject to frequent failure (Fig. 15b) | (Chen et al., 2025a, Table 7; Sec. 4.2.2) |
Kohlbrecher et al., 2011 · Text Sec. V 本方法 1 筆
指標navigation filter real-time rate
資料集與序列不適用
表格設定（擷取紀錄原文）：Navigation filter update rate; SLAM pose fused asynchronously (Kohlbrecher et al., 2011, Text Sec. V)
navigation filter real-time rate，不適用
這張表在此指標與資料序列只列出本方法一筆，沒有可並列的其他方法，因此不畫圖，數值與出處見下表。這是 Kohlbrecher et al., 2011 在此表設定下報告的數值（author-reported results），不代表方法在其他資料或設定下的表現。
| 方法（原文寫法） | 報告值 | 出處 |
|---|---|---|
| navigation filter (EKF)本方法原文提出 | 100 Hz | (Kohlbrecher et al., 2011, Sec. V) |
來源
Kohlbrecher et al., 2011
（2011）A flexible and scalable SLAM system with full 3D motion estimation2011 IEEE International Symposium on Safety, Security, and Rescue Robotics (SSRR), pp. 155-160
DOI 10.1109/ssrr.2011.6106777程式碼
同儕審查已出版已讀全文經典查證後修正
相關版本
- 程式碼釋出：hector_slam ROS packages (tu-darmstadt-ros-pkg) https://github.com/tu-darmstadt-ros-pkg/hector_slam
程式碼：https://github.com/tu-darmstadt-ros-pkg/hector_slam（授權：BSD (license tag in hector_mapping/package.xml)）。有公開程式碼不等於已被重現，也不代表目前版本與論文版本相同。