Probabilistic degeneracy detection (DRPM)

Propagates point and normal noise to the point-to-plane Hessian to compute a per-direction degeneracy probability and smoothly attenuates updates accordingly.

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。

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。

作者報告的優勢與限制

優勢

• Parameters taken from the LiDAR datasheet (sigma_p = 1 cm for VLP-16, 1.5 cm for the Ouster sensors) and kept unchanged between experiments 1 and 2 and between experiments 3 and 4 (Sec. IV-B to IV-E)
• Lowest full-trajectory APE in the Seemühle mine: mean 0.04 m (SD 0.02) versus 0.57, 0.18 and 0.78 m for Zhang, Hinduja and Lee (Table II)
• No degeneracy-induced drift in any of the four experiments, whereas Zhang and Lee drifted in all four and Hinduja in two (Table I)
• In the Fyllingsdalen bicycle tunnel the method recovered the roughly 500 m tunnel section length, while tuned Zhang underestimated the longitudinal translation by about 110 m (Sec. IV-E)
• Median 17 ms per scan-matching step on a laptop CPU with 5.4% overhead for detection (Sec. IV-F)

限制

• Experiments 1, 3 and 4 (including the construction site) lack ground truth, so only qualitative results are given there (Sec. IV-A)
• Relies on outlier rejection or robust costs so that the Gaussian noise assumptions on points, normals and residuals hold (Sec. II-C)
• Degeneracy at the construction site and in the tank was induced by cutting the horizontal FOV to 180 degrees; at the construction site the full 360 degree FOV affected scan-to-map ICP only mildly (Sec. IV-B, IV-D)
• Hinduja et al. performed comparably in experiments 2 and 4, and Zhang et al. gave qualitatively comparable results in experiment 2 when tuned per dataset (lambda_min = 150) (Table I, Sec. IV-C)

營建工程相關證據

實驗一在瑞士 Rümlang 工地的大型開闊區域以 ANYmal C（VLP-16，10 Hz）與腿式里程計先驗測試地面平面造成的退化（沿地面平移與繞法向旋轉不受約束）。作者說明使用完整 360 度視野時掃描對地圖 ICP 只受輕微影響，因此把水平視野人為縮為 180 度來研究退化；該場地無真值，只有定性比較（Sec. IV-B, Fig. 2, Table I）。

原文驗證環境：施工中工地、地下或隧道、基礎設施、獨立參考量測

來源

• Hatleskog & Alexis, 2024

  Johan Hatleskog, Kostas Alexis（2024）Probabilistic Degeneracy Detection for Point-to-Plane Error MinimizationIEEE Robotics and Automation Letters, 9(12), pp. 11234-11241

  DOI 10.1109/lra.2024.3484153arXiv 2410.10784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：Probabilistic Degeneracy Detection for Point-to-Plane Error Minimization https://arxiv.org/abs/2410.10784

程式碼：https://github.com/ntnu-arl/drpm（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。

回到方法圖鑑