PINGS
PINGS 在 PIN-SLAM 的神經點上同時編碼連續 SDF 與高斯潑濺輻射場,並加上兩者之間的幾何一致性約束,使影像的稠密光度線索回饋改善距離場,距離場則約束高斯分布。作者在 Oxford Spires 以 Leica RTC360 地面雷射掃描(TLS)參考地圖評估表面重建,但該評估關閉定位模組、全部使用真值位姿,因此量到的是建圖元件品質。
本頁內容
Unifies an SDF and a Gaussian radiance field in one neural point map with mutual geometric consistency for LiDAR-visual SLAM.
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 | Ouster OS1-128 LiDAR, 128 beams, 45 deg vertical FOV, 10 Hz, mounted horizontally (in-house car dataset)、four Basler Ace cameras giving 360 deg coverage at 10 Hz (in-house car dataset)、Oxford Spires handheld rig: 64-beam LiDAR and three global-shutter cameras (dataset) |
|---|---|
| 原文測試平台 | vehicle、handheld |
| 狀態估計 | LiDAR odometry by Gauss-Newton alignment of each scan to the SDF zero level set using only SDF values and gradients (no explicit correspondences); camera poses initialised from LiDAR odometry and extrinsics and refined by gradient descent during radiance-field training to absorb imperfect camera and LiDAR synchronization; loop closure detection and pose graph optimization run in parallel |
| 資料關聯 | point-to-implicit SDF for odometry; photometric Gaussian-splatting loss with SDF-radiance geometric consistency for mapping |
| 時間表示 | discrete poses |
| 去畸變 | 原文未報告 |
| 迴圈閉合 | loop closure detection running in parallel (inherited from PIN-SLAM) |
| 全域最佳化 | pose graph optimization |
| 地圖表示 | neural points jointly encoding a signed distance field and a Gaussian-splatting radiance field |
| 先驗資訊 | none (per-scene online optimization, no pre-trained priors) |
| 可輸出幾何 | SDF mesh via marching cubes; rendered RGB and depth from the Gaussian radiance field |
| 計算需求 | SDF mapping and LiDAR odometry at sensor rate, but overall ~5 s per frame on an NVIDIA A6000 due to radiance-field mapping (Sec. V) |
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 | Ouster OS1-128 | 方法輸入 | in-house car dataset | 128 beams, 45 deg vertical FOV, 10 Hz, mounted horizontally on a robot car | (Pan et al., 2025, Sec. IV-A1) |
| LiDAR | 64-beam LiDAR (model not named) | 資料集感測器 | Oxford Spires | on a handheld system | (Pan et al., 2025, Sec. IV-A1) |
| 地面雷射掃描儀(TLS) | geo-referenced terrestrial laser scans (scanner model not named) | 參考或真值量測 | in-house car dataset | precise constraints in the offline reference-pose bundle adjustment | (Pan et al., 2025, Sec. IV-A1) |
| 地面雷射掃描儀(TLS) | Leica RTC360 | 參考或真值量測 | Oxford Spires | millimetre-accurate reference map | (Pan et al., 2025, Sec. IV-A1) |
| GNSS 接收器 | RTK-GNSS (model not named) | 參考或真值量測 | in-house car dataset | incorporated into offline LiDAR bundle adjustment for reference poses | (Pan et al., 2025, Sec. IV-A1) |
| 相機 | Basler Ace (four units) | 方法輸入 | in-house car dataset | 360 deg visual coverage, 10 Hz; images used at 512 x 1,032 | (Pan et al., 2025, Sec. IV-A1; Sec. IV-A2) |
| 相機 | three global-shutter cameras (model not named) | 資料集感測器 | Oxford Spires | images used at 540 x 720 | (Pan et al., 2025, Sec. IV-A1; Sec. IV-A2) |
| 載具平台 | robot car | 方法輸入 | in-house car dataset | drives of 5.0 km (about 10,000 scans and 40,000 images) and 3.7 km | (Pan et al., 2025, Sec. IV-A1; Table III) |
| 運算硬體 | NVIDIA A6000歸入:NVidia A6000 | 執行運算平台 | 未標示 | single GPU; overall about 5 s per frame | (Pan et al., 2025, Sec. IV-A2; Sec. V) |
作者報告的優勢與限制
優勢
- Better Chamfer distance and F-score than OpenMVS, Nerfacto, GSS, VDB-Fusion and PIN-SLAM on Oxford Spires (Sec. IV-C; Table II)
- Radiance-field cues improve SDF accuracy over PIN-SLAM at equal resolution (Sec. IV-C)
- Lower drift and ATE than PIN-SLAM on two long in-house car sequences, e.g. ATE 1.99 m vs 3.17 m on the 5.0 km sequence (Table III)
限制
- ~5 s per frame overall on an A6000 (Sec. V)
- No pre-trained priors (Sec. V)
- No explicit 4D (dynamic) modelling (Sec. V)
營建工程相關證據
論文未在營建場域測試;以 TLS 參考的校園資料(Oxford Spires)評估幾何,方法設計(LiDAR 幾何+影像外觀)與既有建物紀錄高度相關(推論)。
原文驗證環境:公開基準、獨立參考量測、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 3 個比較組,合計 25 筆紀錄。
Pan et al., 2025 · Table II 本方法 16 筆
表格設定(擷取紀錄原文):Oxford Spires surface reconstruction against the millimetre-accurate Leica RTC360 TLS reference map; localization disabled and ground-truth poses used for all methods; OpenMVS and Nerfacto results taken from the benchmark (offline batch); meshes at 0.1 m resolution; F-score threshold 0.1 m; precision and recall columns not extracted (Pan et al., 2025, Table II)
Accuracy error,Oxford Spires · Blenheim Palace 05
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., 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(Pan et al., 2025, Table II)
| 方法(原文寫法) | 報告值 | 出處 |
|---|---|---|
| OpenMVS [5] (offline) | 0.126 m | (Pan et al., 2025, Table II) |
| Nerfacto [62] (offline) | 0.302 m | (Pan et al., 2025, Table II) |
| GSS [11] | 0.204 m | (Pan et al., 2025, Table II) |
| VDB-Fusion [67] | 0.098 m | (Pan et al., 2025, Table II) |
| PIN-SLAM [51] | 0.078 m | (Pan et al., 2025, Table II) |
| PINGS (Ours)本方法原文提出 | 0.072 m | (Pan et al., 2025, Table II) |
Pan et al., 2025 · Table III 本方法 8 筆
表格設定(擷取紀錄原文):In-house car dataset (Bonn), full sequences; reference poses from offline LiDAR bundle adjustment with RTK-GNSS, point cloud alignment and geo-referenced TLS constraints; odometry methods above, SLAM methods below; ATE alignment not stated (Pan et al., 2025, Table III)
ARTE [%] (average relative translation error),in-house car dataset · Seq. 1 (5.0 km)
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., 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(Pan et al., 2025, Table III)
| 方法(原文寫法) | 報告值 | 出處 |
|---|---|---|
| F-LOAM [69] | 1.96% | (Pan et al., 2025, Table III) |
| KISS-ICP [68] | 1.49% | (Pan et al., 2025, Table III) |
| PIN odometry [51] | 0.95% | (Pan et al., 2025, Table III) |
| PINGS odometry本方法原文提出 | 0.73% | (Pan et al., 2025, Table III) |
| SuMa [3] | 5.55% | (Pan et al., 2025, Table III) |
| MULLS [50] | 2.23% | (Pan et al., 2025, Table III) |
| PIN-SLAM [51] | 1% | (Pan et al., 2025, Table III) |
| PINGS (Ours)本方法原文提出 | 0.68% | (Pan et al., 2025, Table III) |
Pan et al., 2025 · Text Sec. V 本方法 1 筆
指標overall processing time of around five seconds per frame
資料集與序列in-house car dataset and Oxford Spires (not specified)
表格設定(擷取紀錄原文):Overall processing time stated in the limitations; SDF mapping and LiDAR odometry run at sensor frame rate, radiance-field mapping dominates (Pan et al., 2025, Text Sec. V)
overall processing time of around five seconds per frame,in-house car dataset and Oxford Spires (not specified)
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 Pan et al., 2025 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 |
|---|---|---|
| PINGS (Ours)本方法原文提出硬體:single NVIDIA A6000 GPU | 5 s | (Pan et al., 2025, Sec. V Limitations) |
來源
Pan et al., 2025
(2025)PINGS: Gaussian Splatting Meets Distance Fields within a Point-Based Implicit Neural MapRobotics: Science and Systems XXI
DOI 10.15607/rss.2025.xxi.040arXiv 2502.05752程式碼
同儕審查已出版已讀全文近十年
相關版本
- 預印本:arXiv:2502.05752 https://arxiv.org/abs/2502.05752
程式碼:https://github.com/PRBonn/PINGS(授權:MIT)。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。