FastSLAM 2.0｜SLAM 點雲圖鑑

Adds measurement-informed pose proposals to FastSLAM and proves single-particle convergence for linear-Gaussian SLAM, with accuracy gains reported on Victoria Park.

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。

FastSLAM 2.0 的技術屬性
感測輸入	["range finder (type not stated in the paper)","vehicle odometry (described as relatively inaccurate)","GNSS (DGPS used for evaluation only)"]
原文測試平台	vehicle、simulation
狀態估計	Rao-Blackwellized particle filter with pose proposal conditioned on the current measurement
資料關聯	per-particle maximum-likelihood association that accounts for the sampled pose (Sec. 4.4); new features created when the measurement probability falls below a threshold, and spurious features removed by a log-odds existence filter (Sec. 4.5)
時間表示	discrete poses
去畸變	原文未報告
迴圈閉合	implicit; no explicit module
全域最佳化	none
地圖表示	point landmarks (per-particle EKFs)
先驗資訊	none
可輸出幾何	landmark map and vehicle path
計算需求	each update takes constant time with M = 1 (Sec. 5); on a 1 GHz Pentium PC the Victoria Park data (1,550 s acquisition) took 54 s with FastSLAM 2.0 (M = 1), 315 s with regular FastSLAM (M = 50) and 7,807 s with an EKF (Sec. 6 table)

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。

原文使用的設備
類別	型號（原文寫法）	角色	資料集	原文規格	出處
GNSS 接收器	differential GPS	參考或真值量測	Victoria Park (Sydney)	used for evaluation only	(Montemerlo et al., 2003, Sec. 6)
載具平台	outdoor vehicle	資料集感測器	Victoria Park (Sydney)	原文未報告	(Montemerlo et al., 2003, Sec. 6)
運算硬體	1GHz Pentium PC	執行運算平台	未標示	1 GHz	(Montemerlo et al., 2003, Sec. 6 (runtime table))
其他	range finder	資料集感測器	Victoria Park (Sydney)	low-noise; used for landmark detection (type not stated)	(Montemerlo et al., 2003, Sec. 6)
其他	odometry	資料集感測器	Victoria Park (Sydney)	described as relatively inaccurate; raw-odometry average RMS error 93.6 m	(Montemerlo et al., 2003, Sec. 6, Fig. 1a)

作者報告的優勢與限制

優勢

Convergence proof for linear SLAM with M = 1 particle, described by the authors as the first for a constant-time SLAM algorithm (Sec. 5
Sec. 7) | Abstract reports an order-of-magnitude accuracy improvement over the original FastSLAM on real-world data
Fig. 2 shows FastSLAM 2.0 performing about equally well for any particle number while regular FastSLAM performs poorly with very small particle sets (abstract
Sec. 6) | With M = 1 on Victoria Park, accuracy previously reached only by O(N^2) EKF-style methods or FastSLAM with M = 50
processing took 54 s, under 4% of the 1,550 s acquisition time (Sec. 6)

限制

Authors state multiple particles remain warranted when data association is ambiguous (Sec. 7).
The proof applies only to a restricted linear Gaussian SLAM class (Sec. 5).

營建工程相關證據

原文未報告（僅使用 Victoria Park 公園的戶外車輛資料與模擬資料，與營建場域無關）

原文驗證環境：公開基準、模擬、獨立參考量測

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 2 個比較組，合計 3 筆紀錄。

Montemerlo et al., 2003 · Text Sec. 6 本方法 2 筆

指標number of landmarks in the map

資料集與序列Victoria Park (Sydney) · full data set (3.5 km)

表格設定（擷取紀錄原文）：Number of landmarks in the Victoria Park map with M = 1, without and with the feature-management rule of Sec. 4.5 (Montemerlo et al., 2003, Text Sec. 6)

number of landmarks in the map，Victoria Park (Sydney) · full data set (3.5 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。

這些是 Montemerlo et al., 2003 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。

統計量：原文未報告；對齊方式：未對齊；單位：landmarks；場景：outdoor park

資料來源作者報告值(Montemerlo et al., 2003, Text Sec. 6)

數值與出處
方法（原文寫法）	報告值	出處
FastSLAM 2.0, M = 1, without feature management本方法原文提出	768 landmarks	(Montemerlo et al., 2003, Sec. 6, Fig. 1b)
FastSLAM 2.0, M = 1, with feature management本方法原文提出	343 landmarks	(Montemerlo et al., 2003, Sec. 6, Fig. 1c)

Montemerlo et al., 2003 · Table Sec. 6 本方法 1 筆

指標processing time for the whole data set

資料集與序列Victoria Park (Sydney) · full data set (3.5 km)

表格設定（擷取紀錄原文）：Total time to process the Victoria Park data set on a 1 GHz Pentium PC; data acquisition took 1,550 s (Montemerlo et al., 2003, Table Sec. 6)

processing time for the whole data set，Victoria Park (Sydney) · full data set (3.5 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。

這些是 Montemerlo et al., 2003 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。

統計量：原文未報告；對齊方式：未對齊；單位：s；場景：outdoor park

資料來源作者報告值(Montemerlo et al., 2003, Table Sec. 6)

數值與出處
方法（原文寫法）	報告值	出處
EKF硬體：1GHz Pentium PC	7807 s	(Montemerlo et al., 2003, Sec. 6 table)
regular FastSLAM, M=50 particles硬體：1GHz Pentium PC	315 s	(Montemerlo et al., 2003, Sec. 6 table)
FastSLAM 2.0, M=1 particle本方法原文提出硬體：1GHz Pentium PC	54 s	(Montemerlo et al., 2003, Sec. 6 table)

來源

Montemerlo et al., 2003
Michael Montemerlo, Sebastian Thrun, Daphne Koller, Ben Wegbreit（2003）FastSLAM 2.0: An Improved Particle Filtering Algorithm for Simultaneous Localization and Mapping that Provably ConvergesProceedings of the Eighteenth International Joint Conference on Artificial Intelligence (IJCAI-03), pp. 1151-1156
來源頁面
同儕審查已出版已讀全文經典查證後修正

回到方法圖鑑