KISS-SLAM
KISS-SLAM 將 KISS-ICP 延伸為完整 LiDAR-only SLAM：依行進距離切分局部地圖，以關鍵位姿作為位姿圖節點。迴圈偵測將局部地圖地面對齊後投影成鳥瞰密度影像，以 ORB 描述子比對，再以 3D 配準與重疊率（門檻 40%）驗證後加入位姿圖最佳化。處理完成後另做離線細粒度位姿圖最佳化，將殘餘漂移分配到局部軌跡。
本頁內容
Extends KISS-ICP to LiDAR-only SLAM with distance-based local maps, BEV density-image ORB loop detection verified by 3D overlap, and pose-graph optimization, emphasizing minimal tuning.
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 | 3D LiDAR only |
|---|---|
| 原文測試平台 | vehicle、wheeled UGV、handheld |
| 狀態估計 | KISS-ICP odometry; pose graph over local-map keyposes; offline fine-grained pose graph over scan poses after processing |
| 資料關聯 | point-to-point ICP (odometry); loop verification by registration of voxel mean-and-normal clouds |
| 時間表示 | discrete poses with constant-velocity deskew (KISS-ICP) |
| 去畸變 | constant-velocity per-point deskew inherited from KISS-ICP |
| 迴圈閉合 | ground alignment, bird's-eye-view density images, ORB descriptors with database search, RANSAC 2D alignment, then 3D registration and overlap check (accepted above 40%) |
| 全域最佳化 | pose graph optimization of local-map keyposes on accepted closures; final offline fine-grained PGO redistributing drift within local maps |
| 地圖表示 | keypose-anchored local maps (voxel grids) split by travelled distance; output 3D occupancy grid |
| 先驗資訊 | none |
| 可輸出幾何 | globally corrected trajectory, local point maps and 3D occupancy grid (0.05 m voxels in navigation experiment) |
| 計算需求 | faster than sensor frame rate on robot Intel NUC (i7, 32 GB RAM) (Sec. IV-D) |
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 | Aeva | 資料集感測器 | HeLiPR | HeLiPR sensor with different ranging technology and scan pattern | (Guadagnino et al., 2025a, Sec. IV-A; Table III; Table VI caption) |
| LiDAR | Avia歸入：Livox Avia | 資料集感測器 | HeLiPR | HeLiPR sensor; non-repetitive pattern shown in Fig. 1 | (Guadagnino et al., 2025a, Fig. 1; Table III) |
| LiDAR | Ouster | 資料集感測器 | HeLiPR | HeLiPR scanner, called 'the Ouster scanner'; model not stated | (Guadagnino et al., 2025a, Fig. 1; Table III) |
| LiDAR | Velodyne VLP-16 | 資料集感測器 | HeLiPR | excluded because of self-occlusion by surrounding sensors | (Guadagnino et al., 2025a, Sec. IV-A) |
| LiDAR | Hesai XT-32 | 方法輸入 | 未標示 | 3D LiDAR used for mapping; max range processed reduced to 50 m indoors | (Guadagnino et al., 2025a, Sec. IV-D; Fig. 3) |
| LiDAR | SICK TiM781S | 比較對象設備 | 未標示 | 2D LiDAR mounted 0.16 m above ground on the Dingo | (Guadagnino et al., 2025a, Sec. IV-D; Fig. 3) |
| 相機 | 原文未報告 | 參考或真值量測 | 未標示 | upward-looking camera detecting AprilTags on the office ceiling to give ground-truth poses; model not stated | (Guadagnino et al., 2025a, Sec. IV-D) |
| 載具平台 | Clearpath Husky | 方法輸入 | 未標示 | mapping robot carrying the Hesai XT-32 | (Guadagnino et al., 2025a, Sec. IV-D; Fig. 3) |
| 載具平台 | Clearpath Dingo | 比較對象設備 | 未標示 | second robot localized on the sliced 2D map; also recorded data for the GMapping baseline map | (Guadagnino et al., 2025a, Sec. IV-D; Fig. 3) |
| 運算硬體 | Intel NUC | 執行運算平台 | 未標示 | Intel i7 processor, 32 GB RAM; KISS-SLAM ran faster than the sensor frame rate on board | (Guadagnino et al., 2025a, Sec. IV-D) |
作者報告的優勢與限制
優勢
- Same parameter configuration across MulRan, HeLiPR (Aeva, Avia and Ouster; VLP-16 excluded), Apollo, NCLT and Newer College (Sec. IV-A)
- Only method with a result in every reported sequence; zero parameter changes across HeLiPR sensors and Newer College, versus 7 to 25 for PIN-SLAM, SuMa and CT-ICP; MULLS needed 0 changes between HeLiPR Avia and Ouster and 25 from Ouster to NCD-2020, with no working parameter set for pairs involving Aeva (Sec. IV-B; Sec. IV-C; Table VI)
- Lower ATE than KISS-ICP odometry on all MulRan scenes, e.g. Sejong 178.88 m versus 316.20 m (version-of-record Table II)
- Map output usable for 2D occupancy-grid global localization on a real robot; authors report no significant difference from a GMapping map, although pose-tracking ATE RMS was higher in the dynamic sequences (Sec. IV-D; Table VII)
限制
- Velodyne VLP-16 HeLiPR sequences excluded due to self-occlusion (Sec. IV-A)
- Maximum range reduced from 100 m to 50 m for the office navigation experiment (Sec. IV-D)
- Highest ATE among methods that ran on Newer College 2021 stairs (3.58 m) and 2021 cloister (0.40 m) (Table V)
- Large absolute errors remain on HeLiPR Bridge (Aeva 98.61 m, Avia 148.88 m) and MulRan Sejong (178.88 m) (Tables II-III)
- Loop closure relies on ground alignment and BEV projection (inference: may be less suited to multi-storey interiors without distinct floor layouts)
營建工程相關證據
原文未報告（評估為道路、校園與辦公室環境；辦公室僅用於導航定位測試）
原文驗證環境：公開基準、受控實驗
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 6 個比較組，合計 33 筆紀錄。以下列出本方法紀錄最多的 4 組，其餘 2 組列在最後，並連到性能比較頁。
Guadagnino et al., 2025a · Table III 本方法 9 筆
指標ATE [m] (evo)
表格設定（擷取紀錄原文）：ATE from evo; the paired relative KITTI metric (%) is omitted; '-' means the run failed because errors exceeded a sequence-specific threshold (version of record); same KISS-SLAM configuration for all runs; values averaged over three runs per scene (Guadagnino et al., 2025a, Table III)
ATE [m] (evo)，HeLiPR · Bridge Aeva
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。
- 失敗
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 Guadagnino et al., 2025a 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(Guadagnino et al., 2025a, Table III)
| 方法（原文寫法） | 報告值 | 出處 |
|---|---|---|
| PIN-SLAM | 無數值失敗註記（擷取紀錄）：failed ('-': error exceeded a sequence-specific threshold) | (Guadagnino et al., 2025a, Table III) |
| SuMa | 無數值失敗註記（擷取紀錄）：failed ('-': error exceeded a sequence-specific threshold) | (Guadagnino et al., 2025a, Table III) |
| CT-ICP | 無數值失敗註記（擷取紀錄）：failed ('-': error exceeded a sequence-specific threshold) | (Guadagnino et al., 2025a, Table III) |
| MULLS | 356.06 m | (Guadagnino et al., 2025a, Table III) |
| Ours (KISS-SLAM)本方法原文提出 | 98.61 m | (Guadagnino et al., 2025a, Table III) |
Guadagnino et al., 2025a · Table IV 本方法 7 筆
指標ATE [m] (evo)
表格設定（擷取紀錄原文）：ATE from evo; the paired relative KITTI metric (%) is omitted; '-' means the run failed because errors exceeded a sequence-specific threshold (version of record); same KISS-SLAM configuration for all runs (Guadagnino et al., 2025a, Table IV)
ATE [m] (evo)，Apollo · BTS 2018-10-12
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 Guadagnino et al., 2025a 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(Guadagnino et al., 2025a, Table IV)
| 方法（原文寫法） | 報告值 | 出處 |
|---|---|---|
| PIN-SLAM | 6.1 m | (Guadagnino et al., 2025a, Table IV) |
| SuMa | 181.19 m | (Guadagnino et al., 2025a, Table IV) |
| CT-ICP | 10.81 m | (Guadagnino et al., 2025a, Table IV) |
| MULLS | 104.14 m | (Guadagnino et al., 2025a, Table IV) |
| Ours (KISS-SLAM)本方法原文提出 | 3.74 m | (Guadagnino et al., 2025a, Table IV) |
Guadagnino et al., 2025a · Table V 本方法 7 筆
指標ATE [m] (evo)
表格設定（擷取紀錄原文）：ATE from evo; the paired relative KITTI metric (%) is omitted; '-' means the run failed because errors exceeded a sequence-specific threshold (version of record); same KISS-SLAM configuration for all runs (Guadagnino et al., 2025a, Table V)
ATE [m] (evo)，Newer College · 2020 01-short
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 Guadagnino et al., 2025a 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(Guadagnino et al., 2025a, Table V)
| 方法（原文寫法） | 報告值 | 出處 |
|---|---|---|
| PIN-SLAM | 0.42 m | (Guadagnino et al., 2025a, Table V) |
| SuMa | 2.06 m | (Guadagnino et al., 2025a, Table V) |
| CT-ICP | 0.63 m | (Guadagnino et al., 2025a, Table V) |
| MULLS | 0.47 m | (Guadagnino et al., 2025a, Table V) |
| Ours (KISS-SLAM)本方法原文提出 | 0.3 m | (Guadagnino et al., 2025a, Table V) |
Guadagnino et al., 2025a · Table VII 本方法 5 筆
指標ATE translation RMS [cm] (mean over 10 runs)
表格設定（擷取紀錄原文）：2D Monte-Carlo localization (RVP-Loc, Clearpath Dingo with SICK TiM781S) on a 2D map sliced from the KISS-SLAM 3D occupancy grid versus a GMapping map; pose-tracking ATE translation RMS, mean of 10 runs; ground truth from ceiling AprilTags seen by an upward camera; success rate and convergence time columns omitted (Guadagnino et al., 2025a, Table VII)
ATE translation RMS [cm] (mean over 10 runs)，authors' office sequences · Static Sequence 1
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 Guadagnino et al., 2025a 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(Guadagnino et al., 2025a, Table VII)
| 方法（原文寫法） | 報告值 | 出處 |
|---|---|---|
| GMapping map | 9.48 cm | (Guadagnino et al., 2025a, Table VII) |
| Ours (KISS-SLAM map)本方法原文提出 | 9.71 cm | (Guadagnino et al., 2025a, Table VII) |
其他比較組
來源
Guadagnino et al., 2025a
（2025）KISS-SLAM: A Simple, Robust, and Accurate 3D LiDAR SLAM System With Enhanced Generalization Capabilities2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 5363-5370
DOI 10.1109/iros60139.2025.11246613arXiv 2503.12660程式碼
同儕審查已出版已讀全文近十年查證後修正
相關版本
- 預印本：KISS-SLAM (arXiv v1) https://arxiv.org/abs/2503.12660
- 程式碼釋出：PRBonn/kiss-slam https://github.com/PRBonn/kiss-slam
程式碼：https://github.com/PRBonn/kiss-slam（授權：MIT (LICENSE file checked)）。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。