Pith. sign in

REVIEW 4 major objections 5 minor 28 references

FDIT* claims up to 34.65% lower initial solution cost than EIT* in high-dimensional benchmarks.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

FDIT* uses repulsive forces from invalid samples and an elliptical k-nearest-neighbor search to guide sampling-based planning, reporting up to 34.65% lower initial path cost than EIT* on R4-R16 benchmarks.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection Genuinely useful heuristic extension to EIT* with plausible empirical gains, but the asymptotic-optimality claim is unsupported and the formal section is largely filler. the 4 major comments →

arxiv 2508.19771 v1 pith:UB2B3ADD submitted 2025-08-27 cs.RO

Elliptical K-Nearest Neighbors -- Path Optimization via Coulomb's Law and Invalid Vertices in C-space Obstacles

Liding Zhang , Zhenshan Bing , Yu Zhang , Kuanqi Cai , Lingyun Chen , Fan Wu , Sami Haddadin , Alois Knoll This is my paper
classification cs.RO
keywords sampling-based motion planningk-nearest neighborsinvalid verticesCoulomb's lawelliptical searchhigh-dimensional planninganytime optimal planningmobile manipulation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

FDIT* is a sampling-based path planner that tries to make better use of the samples that land inside obstacles. It treats each invalid sample as a repulsive electric charge and each valid sample as an attractive charge, then uses the resulting force vector to stretch the planner's nearest-neighbor search region into an ellipsoid pointing toward promising free space. The paper claims this elliptical k-nearest-neighbors search, layered on top of the EIT* planner, produces cheaper initial paths and faster convergence in high-dimensional, confined C-spaces. On dividing-wall and random-rectangle benchmarks from R4 to R16, FDIT* reports up to 34.65% lower median initial solution cost than EIT*, and a higher success rate on a real 8-DoF mobile-manipulator insertion task. The authors also claim FDIT* is asymptotically optimal because it inherits that property from EIT*, though they do not prove the deformed neighbor graph preserves the required conditions.

Core claim

The central discovery is that invalid vertices—samples that fall inside C-space obstacles—carry useful directional information and can be repurposed to steer nearest-neighbor selection. By assigning repulsive Coulomb-like charges to invalid samples and attractive charges to valid ones, and summing these forces on the current vertex, the algorithm obtains a force direction. That direction scales each axis of an ellipsoidal distance metric, defining an elliptical k-nearest-neighbor search that includes nearby valid vertices aligned with the force while excluding those behind obstacles. This changes the graph-rewiring and search-expansion behavior of EIT*: instead of symmetric spherical neighbo

What carries the argument

The key machinery is a force-directed elliptical k-nearest-neighbors search. It has three parts: (1) a high-dimensional extension of Coulomb's law that defines the force vector between the current state and any valid or invalid sample as proportional to the charge product divided by distance to the (n−1)th power; (2) an iterative procedure that computes a resultant force direction from the current neighbor set, re-filters the neighbor set through an ellipsoidal distance metric whose axis scales are set by that direction, and repeats until the share of invalid samples falls below a threshold; and (3) the standard logarithmic neighbor count k(q), kept from prior k-NN theory. The force directio

Load-bearing premise

The plan assumes that changing how a vertex picks its neighbors to an ellipsoidal, force-directed set does not break the property that random geometric graphs connect well enough to contain near-optimal paths; this is asserted rather than proved.

What would settle it

On the DW-R16 benchmark, run FDIT* with the force-direction computation disabled (fixed spherical k-NN with identical per-vertex effort). If the median initial cost is not meaningfully higher than the force-directed version, the reported 34.65% gain comes from extra computation, not from directional awareness. Separately, construct a narrow-gap problem where all nearby invalid samples lie on one side of the current vertex; if the force direction pushes the ellipse away from the only free gap and the planner fails while standard k-NN succeeds, the deformed neighborhood is not reliably connected

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • For any sampling-based planner that rewires a tree, replacing symmetric k-NN with force-shaped neighbor selection can reduce the cost of the first found solution and the time to reach a given cost level, without changing the underlying sampling distribution.
  • The improvement grows with dimension on the tested benchmarks, from roughly 15% cost reduction in R4 to 34.65% in R16, suggesting the method is especially valuable in high-dimensional C-spaces where narrow passages dominate.
  • In time-bounded applications, such as the 1-second mobile-manipulation insertion task, faster initial solutions translate directly into a higher task success rate under real-time constraints.
  • Because the force direction is computed only from already-sampled vertices, the approach requires no extra oracle or precomputation beyond standard collision checking, so it can be dropped into existing planners with modest overhead.
  • If the asymptotic-optimality inheritance holds, FDIT* remains an anytime optimal planner, so users get both better initial paths and a theoretical convergence guarantee.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • An unproven link between elliptical-KNN and random-geometric-graph connectivity is the main theoretical risk: if the deformed ellipsoidal neighbor graph is not connected with high probability at the same radius or degree parameters, the asymptotic-optimality claim may fail while the empirical cost improvements still stand.
  • The Coulomb analogy suggests a natural extension: give charges magnitudes proportional to how much a valid sample lowered the current best path cost, or how close an invalid sample lies to the free-space boundary, rather than uniform ±1 charges; the paper itself flags this quantification as future work.
  • The same invalid-vertex-as-repulsive-charge idea could be applied to the sampling distribution itself, biasing future samples away from known collision regions, which would compound the effect seen here.
  • One testable prediction is that elliptical-KNN's benefit should increase in environments with many narrow passages and decrease in open, obstacle-sparse spaces; running FDIT* with and without force deformation on both problem types would verify this.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes FDIT*, an extension of the sampling-based optimal planner EIT*, in which the standard nearest-neighbor primitive is replaced by an elliptical k-nearest-neighbors search. The ellipse is oriented and scaled according to a resultant "Coulomb force" computed from nearby valid vertices (attractive) and invalid vertices inside obstacles (repulsive). The paper claims three contributions: an elliptical-KNN search that incorporates invalid vertices, improved path quality with up to a 34.65% reduction in median initial solution cost versus EIT*, and a real-world validation on an 8-DOF DARKO mobile manipulator. The experiments compare FDIT* against RRT-Connect, Informed RRT*, BIT*, AIT*, ABIT*, and EIT* in six benchmark settings in R4, R8, and R16, each run 100 times. The paper also asserts that FDIT* is almost-surely asymptotically optimal and proves a theorem on dynamic k selection.

Significance. The core idea is practically appealing: using invalid samples to shape the neighbor-search region is a simple, transferable mechanism that could improve convergence in obstacle-rich, high-dimensional spaces. The experimental infrastructure is a strength: the method is evaluated with standard OMPL planners, across R4-R16, with 100 runs per task, and on a real robot. If the algorithmic description were complete and the empirical claims had proper statistical support, this would be a useful contribution to sampling-based motion planning. However, the theoretical optimality claim is currently an unsupported one-sentence assertion, the central force-update pseudocode is not fully specified, and the headline cost improvements lack confidence intervals or significance tests.

major comments (4)
  1. [Section V-A] The claim of almost-sure asymptotic optimality is not justified. The paper states: 'Considering EIT* is an almost-surely asymptotically optimal planner [24]. Therefore, FDIT*'s RGG approximation almost certainly contains an optimal path.' This is not a valid inference. FDIT* alters the graph-construction rule: Algorithms 1-3 filter the neighbor set with an ellipse that can exclude valid vertices, so the resulting random geometric graph is not the same object as EIT*'s. No lemma verifies the Karaman-Frazzoli conditions (r-ball connectivity, near-optimal path containment) for the modified graph. Moreover, the cited reference [24] is the Planner Developer Tools paper, not the EIT* optimality proof; the actual EIT* reference is [13]. Since asymptotic optimality is a headline contribution, the paper must either supply a proof or remove the claim and present FDIT* as a heuristic with empirical
  2. [Section IV-C, Theorem 1] Theorem 1 is not a proof. Inequality (8) merely restates the definition of being among the first k elliptical nearest neighbors; no additional content is proved. The proof differentiates k(q) and concludes 'this ensures the search is adaptive and efficient,' which has no logical connection to the stated biconditional. The theorem also does not establish any property needed for asymptotic optimality or completeness. I recommend removing the theorem or replacing it with a clearly stated proposition whose proof directly follows from the ordering of the distances.
  3. [Section IV-C, Algorithm 4] Algorithm 4 is not implementable as written. Lines 6 and 8 accumulate variables F_attractive,i and F_repulsive,i, but line 9 uses an undefined F_i, and no rule is given for how the attractive and repulsive forces are combined. The comment 'add or minus depends on direction' is ambiguous. Additionally, Algorithm 1 accumulates Vvalid in lines 22-23 but returns only VellipseNeighbors, so Vvalid has no effect on the output. These gaps make the core algorithm impossible to reproduce from the text; the pseudocode should be rewritten with explicit vector accumulation and a clear definition of the resultant force.
  4. [Section VI-A, Table I] Table I reports only point estimates of median cost and time. The caption of Fig. 6 promises nonparametric 99% confidence intervals, but Table I does not include them, and the text reports the 15-35% reductions as point estimates. Without confidence intervals or paired significance tests (e.g., Wilcoxon signed-rank over the 100 runs), the claimed improvements could be consistent with sampling noise, particularly in the high-dimensional settings. Please report intervals or tests for the Table I comparisons and clarify how the ten random-rectangle variants are pooled into the RR rows.
minor comments (5)
  1. [Algorithm 1, line 7] The while condition `while Φ ≥ 0.1` appears inconsistent with the surrounding comments. The text says 'skip loop when 90% NN is valid,' which corresponds to stopping when Φ <= 0.1. The comment in the code, 'NN invalid samples less than 10%,' suggests the opposite condition. Please correct this threshold and make the stopping rule explicit.
  2. [Section II-C and Section IV-C, Eqs. (14)-(17)] The paper freely mixes electrostatics terminology: E is called 'electric field intensity' and later 'electromagnetic potential energy,' while several expressions use 1/r^{n-1} for both the field and the force. In n-dimensional electrostatics, the potential scales as 1/r^{n-2} and the force as 1/r^{n-1}. Since this is clearly a heuristic analogy, please state explicitly that the Coulomb law is used by analogy, not as physical field theory.
  3. [Section II-B and Section IV] The symbol q is overloaded: in k(q) it denotes the number of samples, while in Eqs. (3) and (6) it denotes electric charge. This makes the derivation of Eq. (2) harder to follow. Please use, e.g., N for the sample count and q1, q2 for charges.
  4. [Section VI-A, Fig. 6] The text states that unsuccessful runs have infinite costs. If these infinite values are included in the medians or confidence intervals, the curves may be difficult to interpret. Please state how infinite costs are treated in the median cost evolution and in the reported statistics.
  5. [Theorem 1 statement] The phrase 'p satisfies the following condition for a number less than or equal to k(q)' is unclear. Please rephrase as 'p is among the first k(q) vertices when sorted by d_elliptical(x, *),' which is what inequality (8) actually expresses.

Circularity Check

1 steps flagged

No significant circularity: the empirical benchmark comparisons are external and self-contained; one formal theorem restates the elliptical-KNN definition, while the AO claim is an unsupported inheritance rather than a construction-level circularity.

specific steps
  1. self definitional [Section IV-C, Theorem 1 (Eq. 8) and its proof]
    "Theorem 1. (Dynamic selection for elliptical-KNN): Let P be a set of vertices in an n-dimensional space and q a query vertex. Assuming the distance metric is defined by the elliptical equation. The number of nearest neighbors, k, is dynamically determined by the function (2), then a vertex p ∈ P is an elliptical-KNN of q if and only if for all p′ ∈ P , p satisfies the following condition for a number less than or equal to k(q): ∑_{i=1}^n ((q_i-p_i)/v_i)^2 ≤ ∑_{i=1}^n ((q_i-p'_i)/v_i)^2 (8)"

    This 'if and only if' is the definition of being among the k nearest points under the elliptical distance metric d_elliptical from Definition 1: p is in the k-NN set exactly when fewer than k other points have smaller or equal elliptical distance. Equation (2) merely fixes the integer k(q) and plays no role in the equivalence; the proof instead differentiates k(q) and asserts that logarithmic growth 'ensures the search is adaptive and efficient,' then claims the selection is optimized. Thus the theorem does not derive elliptical-KNN from independent conditions—it restates the selection rule by construction. It is presented as formal support for the elliptical-KNN contribution, though it is not the basis of the external benchmark results.

full rationale

The paper's tested performance claims are grounded in external baselines (EIT*, BIT*, ABIT*, AIT*, Informed RRT*, RRT-Connect) on R4-R16 and in a real DARKO robot task, so the cost/success comparisons do not reduce to fitted parameters or to the authors' own definitions. The only definitional circularity found is Theorem 1, which restates the elliptical-kNN selection rule while presenting it as a derived theorem; this is a minor formal flaw rather than a load-bearing support for the empirical results. Separately, Section V-A's asymptotic-optimality argument is not supported: it says 'Considering EIT* is an almost-surely asymptotically optimal planner [24]. Therefore, FDIT*’s RGG approximation almost certainly contains an optimal path,' without a lemma showing that force-direction elliptical-KNN preserves the dense-RGG connectivity and near-optimal-path properties needed for AO. That is a substantive correctness gap and a non-sequitur, but it is a missing proof rather than an equation-level reduction to the paper's inputs, so under the strict circularity standard used here it is not counted as a circular step. No load-bearing self-citation is present; the self-citations that appear are background or future-work references. Overall circularity is low.

Axiom & Free-Parameter Ledger

3 free parameters · 5 axioms · 1 invented entities

The central claim rests on undefined hyperparameters (charges, rho0, v_i) and on an unproven inheritance of asymptotic optimality from EIT*. The virtual force field is an invented modeling device, not an independently evidenced entity. The empirical comparison to external baselines is the main non-circular support.

free parameters (3)
  • charge magnitudes Q_valid and Q_invalid (and k_e)
    Alg. 4 lines 6-8: the resultant force direction depends on the relative magnitudes of attraction from valid vertices and repulsion from invalid vertices. The paper never specifies these values, only noting in Section VII that all vertices currently have the same charge.
  • repulsion range rho0
    Eqs. (12)-(13) introduce rho0 as the range within which an invalid vertex exerts repulsion; Section IV-A says it remains constant, but no value is given, even though it changes which invalid vertices contribute force and thus the search direction.
  • ellipse axis scaling factors v_i
    Definition 1 uses v_i to weight dimensions of the elliptical metric, and Section IV-C says the force component in each dimension weights each axis, but no explicit formula for v_i(F_D) is given, leaving the exact search region undefined.
axioms (5)
  • ad hoc to paper Coulomb's law generalizes to n-dimensional space with inverse (n-1)-power distance for force magnitude
    Eqs. (3) and (6) assert this extension without derivation; it is a heuristic adaptation of a physical law, not a standard result.
  • ad hoc to paper Invalid vertices carry repulsive charge and valid vertices attractive charge of equal magnitude
    Section IV-A and Alg. 4 assign signs and assume equal magnitude; this is a modeling choice with no independent evidence, and the magnitude is left unspecified.
  • domain assumption EIT* is almost-surely asymptotically optimal (cited [24])
    Section V-A relies entirely on this cited property to conclude FDIT* is asymptotically optimal; the argument is an appeal to the predecessor's result.
  • ad hoc to paper The force-direction feedback loop in Alg. 1 terminates with a useful direction
    Alg. 1 loops until the invalid-sample ratio is below 0.1, but no convergence or termination proof is given, and worst-case behavior is not analyzed.
  • standard math Karaman-Frazzoli RGG optimality framework applies
    Section V refers to Definition 24 of [7] to frame the asymptotic-optimality property; the underlying random-geometric-graph theory is standard background.
invented entities (1)
  • Virtual Coulomb/electromagnetic field over C-space samples no independent evidence
    purpose: Guide nearest-neighbor search direction and the orientation of the elliptical search region
    The field is a computational heuristic with no observable consequences outside the planner's behavior; it provides no falsifiable prediction that could be tested independently of the algorithm's performance.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Elliptical K-Nearest Neighbors -- Path Optimization via Coulomb's Law and Invalid Vertices in C-space Obstacles." pith.science (2026). https://pith.science/paper/UB2B3ADD

@misc{pith2026250819771,
  author       = {Pith},
  title        = {Pith review of: Elliptical K-Nearest Neighbors -- Path Optimization via Coulomb's Law and Invalid Vertices in C-space Obstacles},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UB2B3ADD}},
  note         = {Machine review of arXiv:2508.19771}
}
Share X Bluesky LinkedIn Reddit HN
abstract

Path planning has long been an important and active research area in robotics. To address challenges in high-dimensional motion planning, this study introduces the Force Direction Informed Trees (FDIT*), a sampling-based planner designed to enhance speed and cost-effectiveness in pathfinding. FDIT* builds upon the state-of-the-art informed sampling planner, the Effort Informed Trees (EIT*), by capitalizing on often-overlooked information in invalid vertices. It incorporates principles of physical force, particularly Coulomb's law. This approach proposes the elliptical $k$-nearest neighbors search method, enabling fast convergence navigation and avoiding high solution cost or infeasible paths by exploring more problem-specific search-worthy areas. It demonstrates benefits in search efficiency and cost reduction, particularly in confined, high-dimensional environments. It can be viewed as an extension of nearest neighbors search techniques. Fusing invalid vertex data with physical dynamics facilitates force-direction-based search regions, resulting in an improved convergence rate to the optimum. FDIT* outperforms existing single-query, sampling-based planners on the tested problems in R^4 to R^16 and has been demonstrated on a real-world mobile manipulation task.

Figures

Figures reproduced from arXiv: 2508.19771 by Alois Knoll, Fan Wu, Kuanqi Cai, Liding Zhang, Lingyun Chen, Sami Haddadin, Yu Zhang, Zhenshan Bing.

Figure 1
Figure 1. Figure 1: Mobile manipulator robot during a real-time industry stan [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The force directional elliptical k-nearest search is visualized. Black and red dots are sampled as valid and invalid vertices, k donates as the connection parameters, black thin lines as forward rewire trees, magenta lines as the initial path, arrows indicate force directions and orange ellipses denote the area of elliptical-KNN in the forward search. Undesirable vertices are excluded from the neighbors, a… view at source ↗
Figure 3
Figure 3. Figure 3: Illustration of the progression of the search area refinement in FDIT* on a narrow corridor benchmark scenario, it demonstrates how [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Illustrates the simulation (a) and the real-world scenarios of DARKO robot for the industry shelf-arrangement task, (b) shows the [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Simulated planning problems were visualized using a 2D [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 1
Figure 1. Figure 1: Top: Percentage of runs that found a solution at any given time with a Clopper￾Pearson (nonparametric) 99% confidence interval. Bottom: Median cost evolution and median of initial solution with nonparametric 0% confidence intervals. 1 (a) Dividing Wall-gaps (DW) in R4 - MaxTime: 0.3s (c) Dividing Wall-gaps (DW) in R8 - MaxTime: 0.6s (e) Dividing Wall-gaps (DW) in R16 - MaxTime: 1.0s (b) Random Rectangles (… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

28 extracted references · 28 canonical work pages

  1. [1]

    Sampling- based path planning in highly dynamic and crowded pedestrian flow,

    K. Cai, W. Chen, D. Dugas, R. Siegwart, and J. J. Chung, “Sampling- based path planning in highly dynamic and crowded pedestrian flow,” IEEE Transactions on Intelligent Transportation Systems , 2023. I

    work page 2023

  2. [2]

    Penrose, Random geometric graphs

    M. Penrose, Random geometric graphs . OUP Oxford, 2003, vol. 5. I, 4

    work page 2003

  3. [3]

    A formal basis for the heuristic determination of minimum cost paths,

    P. E. Hart, N. J. Nilsson, and B. Raphael, “A formal basis for the heuristic determination of minimum cost paths,” IEEE transactions on Systems Science and Cybernetics , vol. 4, no. 2, pp. 100–107, 1968. I

    work page 1968

  4. [4]

    Lifelong planning a*,

    S. Koenig, M. Likhachev, and D. Furcy, “Lifelong planning a*,” Artificial Intelligence, vol. 155, no. 1-2, pp. 93–146, 2004. I

    work page 2004

  5. [5]

    Randomized kinodynamic plan- ning,

    S. M. LaValle and J. J. Kuffner Jr, “Randomized kinodynamic plan- ning,” The international journal of robotics research , vol. 20, no. 5, pp. 378–400, 2001. I

    work page 2001

  6. [6]

    Rrt-connect: An efficient approach to single-query path planning,

    J. Kuffner and S. LaValle, “Rrt-connect: An efficient approach to single-query path planning,” in Proceedings 2000 ICRA. Millennium Conference. IEEE International Conference on Robotics and Automa- tion. Symposia Proceedings (Cat. No.00CH37065) , vol. 2, 2000, pp. 995–1001 vol.2. I

    work page 2000

  7. [7]

    Sampling-based algorithms for optimal motion planning,

    S. Karaman and E. Frazzoli, “Sampling-based algorithms for optimal motion planning,” The international journal of robotics research , vol. 30, no. 7, pp. 846–894, 2011. I, II, 4, V

    work page 2011

  8. [8]

    Fast marching tree: A fast marching sampling-based method for optimal motion planning in many dimensions,

    L. Janson, E. Schmerling, A. Clark, and M. Pavone, “Fast marching tree: A fast marching sampling-based method for optimal motion planning in many dimensions,” The International Journal of Robotics Research, vol. 34, no. 7, pp. 883–921, 2015. I

    work page 2015

  9. [9]

    Informed rrt*: Optimal sampling-based path planning focused via direct sampling of an admissible ellipsoidal heuristic,

    J. D. Gammell, S. S. Srinivasa, and T. D. Barfoot, “Informed rrt*: Optimal sampling-based path planning focused via direct sampling of an admissible ellipsoidal heuristic,” in 2014 IEEE/RSJ international conference on intelligent robots and systems . IEEE, 2014. I

    work page 2014

  10. [10]

    Informed sampling for asymptotically optimal path planning,

    J. D. Gammell, T. D. Barfoot, and S. S. Srinivasa, “Informed sampling for asymptotically optimal path planning,” IEEE Transactions on Robotics, vol. 34, no. 4, pp. 966–984, 2018. I

    work page 2018

  11. [11]

    Batch informed trees (bit*): Informed asymptotically optimal anytime search,

    J. D. Gammell, T. D. Barfoot, and S. S. Srinivasa, “Batch informed trees (bit*): Informed asymptotically optimal anytime search,” The International Journal of Robotics Research , vol. 39, no. 5, 2020. I

    work page 2020

  12. [12]

    Advanced bit (abit): Sampling- based planning with advanced graph-search techniques,

    M. P. Strub and J. D. Gammell, “Advanced bit (abit): Sampling- based planning with advanced graph-search techniques,” in 2020 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2020, pp. 130–136. I

    work page 2020

  13. [13]

    Adaptively informed trees (ait*) and effort informed trees (eit*): Asymmetric bidirectional sampling- based path planning,

    M. P. Strub and J. D. Gammell, “Adaptively informed trees (ait*) and effort informed trees (eit*): Asymmetric bidirectional sampling- based path planning,” The International Journal of Robotics Research, vol. 41, no. 4, pp. 390–417, 2022. I, II-A

    work page 2022

  14. [14]

    Adaptively informed trees (ait*): Fast asymptotically optimal path planning through adaptive heuristics,

    M. P. Strub and J. D. Gammell, “Adaptively informed trees (ait*): Fast asymptotically optimal path planning through adaptive heuristics,” in 2020 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2020, pp. 3191–3198. I

    work page 2020

  15. [15]

    Tree- based grafting approach for bidirectional motion planning with local subsets optimization,

    L. Zhang, Y . Ling, Z. Bing, F. Wu, S. Haddadin, and A. Knoll, “Tree- based grafting approach for bidirectional motion planning with local subsets optimization,” IEEE Robotics and Automation Letters, vol. 10, no. 6, pp. 5815–5822, 2025. I

    work page 2025

  16. [16]

    Collision detection or nearest-neighbor search? on the computational bottleneck in sampling- based motion planning,

    M. Kleinbort, O. Salzman, and D. Halperin, “Collision detection or nearest-neighbor search? on the computational bottleneck in sampling- based motion planning,” Springer Proceedings in Advanced Robotics , p. 624–639, 2020. I

    work page 2020

  17. [17]

    The number of neighbors needed for con- nectivity of wireless networks,

    F. Xue and P. Kumar, “The number of neighbors needed for con- nectivity of wireless networks,” Wireless Networks, vol. 10, no. 2, p. 169–181, Mar. 2004. II

    work page 2004

  18. [18]

    Random plane networks,

    E. N. Gilbert, “Random plane networks,” Journal of the Society for Industrial and Applied Mathematics , vol. 9, no. 4, pp. 533–543, 1961. II

    work page 1961

  19. [19]

    The critical radius in sampling-based motion planning,

    K. Solovey and M. Kleinbort, “The critical radius in sampling-based motion planning,” The International Journal of Robotics Research , vol. 39, no. 2-3, pp. 266–285, 2020. II-A

    work page 2020

  20. [20]

    Nearest neighbor search: the old, the new, and the impossible,

    A. Andoni, “Nearest neighbor search: the old, the new, and the impossible,” Ph.D. dissertation, Massachusetts Institute of Technology,

    work page

  21. [21]

    Exact statistical mechanics of a one-dimensional system with coulomb forces,

    A. Lenard, “Exact statistical mechanics of a one-dimensional system with coulomb forces,” Journal of Mathematical Physics , vol. 2, no. 5, pp. 682–693, 1961. II-C

    work page 1961

  22. [22]

    The open motion planning library,

    I. A. Sucan, M. Moll, and L. E. Kavraki, “The open motion planning library,” IEEE Robotics & Automation Magazine , vol. 19, no. 4, pp. 72–82, 2012. II-C

    work page 2012

  23. [23]

    Benchmarking motion planning algorithms: An extensible infrastructure for analysis and visualization,

    M. Moll, I. A. Sucan, and L. E. Kavraki, “Benchmarking motion planning algorithms: An extensible infrastructure for analysis and visualization,” IEEE Robotics & Automation Magazine , vol. 22, no. 3, pp. 96–102, 2015. II-C

    work page 2015

  24. [24]

    Planner developer tools (pdt): Reproducible experiments and statistical analysis for de- veloping and testing motion planners,

    J. D. Gammell, M. P. Strub, and V . N. Hartmann, “Planner developer tools (pdt): Reproducible experiments and statistical analysis for de- veloping and testing motion planners,” in Proceedings of the Workshop on Evaluating Motion Planning Performance (EMPP), IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) ,

    work page

  25. [25]

    A predictive prescription using minimum volume k- nearest neighbor enclosing ellipsoid and robust optimization,

    S. Ohmori, “A predictive prescription using minimum volume k- nearest neighbor enclosing ellipsoid and robust optimization,” Mathe- matics, vol. 9, no. 2, p. 119, 2021. IV-C

    work page 2021

  26. [26]

    Lateral flexion of a compliant spine improves motor performance in a bioinspired mouse robot,

    Z. Bing, A. Rohregger, F. Walter, Y . Huang, P. Lucas, F. O. Morin, K. Huang, and A. Knoll, “Lateral flexion of a compliant spine improves motor performance in a bioinspired mouse robot,” Science Robotics, vol. 8, no. 85, 2023. VII

    work page 2023

  27. [27]

    Human- aware path planning with improved virtual doppler method in highly dynamic environments,

    K. Cai, W. Chen, C. Wang, S. Song, and M. Q.-H. Meng, “Human- aware path planning with improved virtual doppler method in highly dynamic environments,” IEEE Transactions on Automation Science and Engineering, vol. 20, no. 2, pp. 1304–1321, 2022. VII

    work page 2022

  28. [28]

    Estimated informed anytime search for sampling-based planning via adaptive sampler,

    L. Zhang, K. Cai, Y . Zhang, Z. Bing, C. Wang, F. Wu, S. Haddadin, and A. Knoll, “Estimated informed anytime search for sampling-based planning via adaptive sampler,” IEEE Transactions on Automation Science and Engineering , vol. 22, pp. 18 580–18 593, 2025. VII

    work page 2025

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.