SLAMesh
SLAMesh 將掃描點分入體素格,在每格內以高斯過程(Gaussian process)回歸局部表面,於規則分布的位置預測頂點座標與不確定性,再直接連接相鄰頂點形成網格。新掃描同樣重建後,依頂點位置快速建立點對網格(point-to-mesh)對應進行配準;地圖更新只需修正頂點的一維預測值,因此可在 CPU 上即時同時定位與建網格。
本頁內容
Per-cell Gaussian-process surface reconstruction yields regularly placed vertices with uncertainty that are connected into a mesh, enabling fast point-to-mesh registration and incremental mesh updates on a CPU.
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 |
|---|---|
| 原文測試平台 | vehicle、simulation |
| 狀態估計 | iterative point-to-mesh registration with constraint combination (per-layer residual averaging), solved by Levenberg-Marquardt in Ceres with analytic Jacobians; constant-velocity initial guess |
| 資料關聯 | location-based matching of GP-reconstructed vertices to mesh faces in same or adjacent cells, with smoothed face normals |
| 時間表示 | discrete poses |
| 去畸變 | 原文未報告 |
| 迴圈閉合 | no explicit loop closure; optional implicit alignment when registering to revisited areas improves map consistency but was disabled for KITTI odometry metrics (Sec. IV-C) |
| 全域最佳化 | none |
| 地圖表示 | hash map of voxel cells (1.6 m; 1.5 m for the mesh evaluation), each holding up to three Gaussian-process layers of 6 x 6 regularly located vertices with a predicted coordinate and variance, adjacent or diagonal valid vertices connected into triangles |
| 先驗資訊 | none |
| 可輸出幾何 | triangle mesh map with vertex uncertainty |
| 計算需求 | CPU only, 3.6 GHz 8-core Intel i7-11700KF with 8 threads; about 40 Hz on KITTI 07; multithreading cut per-scan time from 57.1 ms to 23.8 ms; with 3 m cells about 80 Hz at 0.88% translation error (Sec. IV-A, IV-D, IV-E) |
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 | Velodyne HDL-64E (simulated)歸入:Velodyne HDL-64E | 資料集感測器 | Mai City | simulated 64-beam LiDAR in CARLA | (Ruan et al., 2023, Sec. IV-B) |
| LiDAR | Velodyne HDL-64E | 資料集感測器 | KITTI odometry | 原文未報告 beyond model | (Ruan et al., 2023, Sec. IV-C) |
| 運算硬體 | Intel i7-11700KF | 執行運算平台 | 未標示 | 3.6 GHz, 8 cores; 8 threads allocated | (Ruan et al., 2023, Sec. IV-A) |
作者報告的優勢與限制
優勢
- On Mai City (d = 0.3 m, each pipeline's own estimated poses) SLAMesh reached F1 80.14% against 76.83% for Voxblox+A-LOAM and 51.08% for Puma (Sec. IV-B, Table I).
- KITTI 00-10 average relative errors 0.676% and 0.291 deg/100 m without loop closure (Sec. IV-C, Table II).
- Multithreading reduced processing from 57.1 ms to 23.8 ms per scan (Sec. IV-E, Fig. 8).
限制
- Mesh-LOAM authors state SLAMesh becomes complicated when modeling complicated geometries (Zhu et al., Mesh-LOAM arXiv v1, Sec. I-II).
- (inference) Fixed cell size and regular vertex locations bound the smallest representable detail.
- (evaluation caveat) KITTI baseline numbers in Table II were imported from the baselines' published papers rather than rerun (Sec. IV-C).
- (evaluation caveat) Voxblox in the A-LOAM+Voxblox pipeline used a 50 m sensor range instead of 100 m, and SLAMesh parameters were tuned manually (Sec. IV-A, IV-D).
營建工程相關證據
僅在 Mai City 模擬與 KITTI 驗證,未涉及營建。
原文驗證環境:模擬、公開基準
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 11 個比較組,合計 82 筆紀錄。以下列出本方法紀錄最多的 4 組,其餘 7 組列在最後,並連到性能比較頁。
Ruan et al., 2023 · Table II 本方法 39 筆
表格設定(擷取紀錄原文):KITTI odometry 00-10, relative translation error; baseline values imported from their published papers (not rerun); SLAMesh without loop closure; w/o Comb. and w/o P2Mesh are ablations (Ruan et al., 2023, Table II)
relative translation error (%),KITTI odometry · 00
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 Ruan et al., 2023 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(Ruan et al., 2023, Table II)
| 方法(原文寫法) | 報告值 | 出處 |
|---|---|---|
| LOAM | 0.78% | (Ruan et al., 2023, Table II) |
| A-LOAM | 0.97% | (Ruan et al., 2023, Table II) |
| Suma | 0.7% | (Ruan et al., 2023, Table II) |
| Suma++ | 0.64% | (Ruan et al., 2023, Table II) |
| Litamin2 | 0.7% | (Ruan et al., 2023, Table II) |
| Puma | 1.46% | (Ruan et al., 2023, Table II) |
| SLAMesh (Ours) Full本方法原文提出 | 0.77% | (Ruan et al., 2023, Table II) |
| SLAMesh w/o Comb.本方法原文提出 | 0.96% | (Ruan et al., 2023, Table II) |
| SLAMesh w/o P2Mesh本方法原文提出 | 0.7% | (Ruan et al., 2023, Table II) |
Zhu et al., 2025 · Table V 本方法 10 筆
表格設定(擷取紀錄原文):Mesh quality with ground-truth poses for all methods, voxel size 0.1 m, settings of SHINE-Mapping; distances in cm; completion ratio and F-score in % at 10 cm (Mai City) and 20 cm (Newer College) (Zhu et al., 2025, Table V)
Comp. (cm), completion,Mai City · Mai City
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 Zhu et al., 2025 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(Zhu et al., 2025, Table V)
| 方法(原文寫法) | 報告值 | 出處 |
|---|---|---|
| VDB Fusion [28] | 6.9 cm | (Zhu et al., 2025, Table V (VoR)) |
| Puma [13] | 32 cm | (Zhu et al., 2025, Table V (VoR)) |
| SHINE-Mapping [30] | 3.2 cm | (Zhu et al., 2025, Table V (VoR)) |
| SLAMesh [14]本方法 | 7.5 cm | (Zhu et al., 2025, Table V (VoR)) |
| Ours原文提出 | 2.5 cm | (Zhu et al., 2025, Table V (VoR)) |
Pan et al., 2024 · Table XI 本方法 8 筆
表格設定(擷取紀錄原文):Newer College 3D reconstruction against the survey-grade TLS reference model (mm-level accuracy); Quad from 02_long, Math Institute from math_easy; 20 cm voxel for all methods; F-score threshold 20 cm; mapping-only methods use KISS-ICP poses; '-' not reported or unavailable (Pan et al., 2024, Table XI)
Map. Acc.,Newer College · Quad (02_long)
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 Pan et al., 2024 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(Pan et al., 2024, Table XI)
| 方法(原文寫法) | 報告值 | 出處 |
|---|---|---|
| VDB-Fusion [77] with KISS-ICP poses | 14.03 cm | (Pan et al., 2024, Table XI) |
| SHINE [100] with KISS-ICP poses | 14.87 cm | (Pan et al., 2024, Table XI) |
| NKSR [20] with KISS-ICP poses | 15.67 cm | (Pan et al., 2024, Table XI) |
| Puma [76] own odometry | 15.3 cm | (Pan et al., 2024, Table XI) |
| SLAMesh [62] own odometry本方法 | 19.21 cm | (Pan et al., 2024, Table XI) |
| Nerf-LOAM [11] own odometry | 12.89 cm | (Pan et al., 2024, Table XI) |
| PIN-SLAM (own odometry)原文提出 | 11.55 cm | (Pan et al., 2024, Table XI) |
Zhu et al., 2025 · Table III 本方法 6 筆
指標ATE (m)
表格設定(擷取紀錄原文):Hilti SLAM Challenge 2021, Ouster OS0-64 data only; all methods run by the authors with their own implementations; caption gives ATE in m while the text says %; most sequences have 3-DoF ground truth; 'x' = failed registration (Zhu et al., 2025, Table III)
ATE (m),Hilti SLAM Challenge 2021 · RPG
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 Zhu et al., 2025 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(Zhu et al., 2025, Table III)
| 方法(原文寫法) | 報告值 | 出處 |
|---|---|---|
| SuMa [8] | 0.262 m | (Zhu et al., 2025, Table III (VoR)) |
| FLOAM [4] | 2.775 m | (Zhu et al., 2025, Table III (VoR)) |
| KISS-ICP [5] | 0.187 m | (Zhu et al., 2025, Table III (VoR)) |
| Puma [13] | 無數值失敗註記(擷取紀錄):failed | (Zhu et al., 2025, Table III (VoR)) |
| SLAMesh [14]本方法 | 0.165 m | (Zhu et al., 2025, Table III (VoR)) |
| Ours原文提出 | 0.173 m | (Zhu et al., 2025, Table III (VoR)) |
其他比較組
來源
Ruan et al., 2023
(2023)SLAMesh: Real-time LiDAR Simultaneous Localization and Meshing2023 IEEE International Conference on Robotics and Automation (ICRA), pp. 3546-3552
DOI 10.1109/icra48891.2023.10161425arXiv 2303.05252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:arXiv:2303.05252 https://arxiv.org/abs/2303.05252
程式碼:https://github.com/lab-sun/SLAMesh(授權:GPL-3.0 (repository LICENSE))。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。