{"id":"54774290-fa14-400b-8c38-476b67072575","arxiv_id":"2501.12417","paper_version":1,"verdict":"REJECT","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"The paper proposes an egoistic, anchorless rigid body localization method, but the translation estimate is unidentifiable and the Nyström completion is invalid for raw distance matrices.","lead":"A new algorithm claims to let a vehicle estimate another vehicle's relative position and orientation from radio distance measurements alone, without knowing the other vehicle's shape. The method relies on multidimensional scaling and matrix completion, but two central equations are mathematically invalid, so the claims are not supported as written.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. (14) applies Nystrom to a raw distance matrix, which is not a PSD kernel and is demonstrably false; the completed D2 corrupts the MDS reconstruction on which both the translation and rotation estimates depend.","rationale":"I read the paper as claiming that from only the first body's shape C1 and the cross-body distances, one can recover the second body's intra-distances D2 and hence the relative translation and rotation, even when N1 and N2 differ and C2 is unknown. For that claim to hold, the only source of information about the target's internal geometry is the Nystrom completion in Eq. (14), so that step is genuinely load-bearing. The paper provides no proof of Eq. (14), and the standard Nystrom theorem does not apply because a raw distance matrix is not a positive semidefinite kernel. The correct completion would operate on the doubly centered Gram matrix, and Eq. (14) is not equivalent to that route. A simple 1D example shows a large quantitative discrepancy, so this is not merely a missing proof or a small approximation error; it changes the reconstructed geometry entirely. The reader's weakest assumption therefore lands, and the rejection verdict stands. I note that I did not reproduce the reader's secondary objection that Eq. (23) is invariant under the translation variable: in a noiseless example the objective is a function of the translation and is minimized at the true value. However, since the fatal flaw at Eq. (14) invalidates the pipeline regardless of that point, the final verdict is unchanged.","tokens_in":943,"tokens_out":1083,"duration_ms":188095,"concrete_test":"Run the noiseless 1D configuration C1 = [-0.5, 0.5], C2 = [10, 12] through Section III: form D1, D12, and true D2; evaluate Eq. (14) and compare its off-diagonal with the true D2. Independently compute the Gram-domain Nystrom completion: K = -1/2 J D^2 J, K22 approx = K21 K11^{-1} K12, then reconstructed squared distances d2_ij = diag(K22)_i + diag(K22)_j - 2 K22_ij, and apply the hollowing operator. If Eq. (14) differs from both the true D2 and the Gram-domain result, the central completion step is false and the downstream estimates in Eqs. (23) and (28) have no valid basis.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Equation (14) is the load-bearing step: it produces the unknown intra-body matrix D2 used in the reconstructed EDM (15) and the MDS embedding (16), and both the translation program (23) and the rotation objective (28) consume that embedding. Standard Nystrom completion applies to positive semidefinite kernel matrices; a raw Euclidean distance matrix is only conditionally negative definite, so the block relation that would justify D2 = H(D12^T D1^{-1} D12) does not exist. The correct completion acts on the doubly centered Gram matrix K = -1/2 J D^2 J, where K22 is approximated by K21 K11^{-1} K12 and squared distances are recovered from K22; Eq. (14) is not algebraically equivalent to that procedure. A noiseless 1D counterexample makes the failure concrete: with C1 = [-0.5, 0.5] and C2 = [10, 12], the true D2 off-diagonal entry is 4, whereas Eq. (14) gives 239.5. Consequently, the reconstructed target points from Eqs. (20)-(21) are not an estimate of the true target shape, and the subsequent translation and rotation estimates (23), (28) are built on an invalid reconstruction. I do not reproduce the reader's separate claim that Eq. (23) is invariant under the translation variable: with the reconstructed distance matrix fixed by the measurements and the point set parameterized as in (22), the objective changes with t in noiseless examples and is minimized at the true translation; the decisive defect is the invalid completion in Eq. (14).","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes an anchorless, egoistic rigid-body localization method in which a rigid body (body 1) estimates the relative translation t and rotation Q of a second rigid body using only cross-body distance measurements, without prior knowledge of the second body's shape. The pipeline in Section III consists of Nyström completion of the unknown intra-body distance matrix D2 in Eq. (14), an MDS embedding in Eqs. (15)-(16), a Procrustes alignment to body 1's frame in Eqs. (19)-(21), an optimization for t in Eq. (23), and a spectral/optimization procedure for Q in Eqs. (24)-(28). The paper reports Monte Carlo RMSE simulations for translation estimation compared with a state-of-the-art method and a genie-aided variant.","tokens_in":10072,"tokens_out":11962,"duration_ms":124988,"significance":"If the central claims were correct, the method would be a useful contribution to V2X perception because it removes the target-shape knowledge requirement common in earlier RBL work. The paper states the problem clearly, and the simulation study is a reasonable first evaluation. However, two load-bearing mathematical steps are not valid as presented: the Nyström completion in Eq. (14) is applied to a matrix that does not satisfy the assumptions of Nyström theory, and the rotation recovery in Eqs. (24) and (28) is subject to a fundamental unidentifiability when the target shape is unknown. The paper supplies no proofs for the completion step, and the simulation results cannot compensate for the invalid reconstruction because the same corrupted matrix feeds the whole estimation pipeline.","major_comments":[{"comment":"Equation (14) applies the Nyström approximation directly to the raw Euclidean distance matrix D. Nyström completion is defined for positive semidefinite kernel matrices; a raw EDM is only conditionally negative definite, so the block relation underlying Eq. (14) does not hold. The correct completion should be applied to the doubly centered Gram matrix K = -1/2 J D^2 J, with K22 approximated by K21 K11^{-1} K12, and squared distances then recovered from K22; Eq. (14) is not algebraically equivalent to that procedure. A noiseless 1D example makes the failure concrete: with C1 = [-0.5, 0.5] and C2 = [10, 12], the true D2 off-diagonal entry is 4, whereas Eq. (14) gives 239.5. Since the completed matrix enters Eqs. (15), (16), and (20)-(21), the reconstructed target points and therefore both the translation estimate (23) and the rotation estimate (28) are built on an invalid reconstruction. Footnote 8's rank condition (rank(D1) >= rank(D2)) is not sufficient; the example satisfies it and still fails catastrophically.","section":"Section III-A, Eq. (14)"},{"comment":"Even if the MDS reconstruction of the target point set were correct, the rotation Q is not identifiable from the assumed measurements when C2 is unknown. Replacing (Q, C2) by (Q R, R^T C2) for any orthogonal R leaves S2 = Q C2 + t unchanged and therefore leaves all measured distances unchanged. The quantities in Eqs. (24) and (27) depend only on Q C2 C2^T Q^T; their eigenvectors are Q times the eigenvectors of C2 C2^T, and without C2 or an additional orientation prior, Q cannot be separated from that product. The permutation ambiguity mentioned in the text for spherical shapes is only one component of this fundamental ambiguity. The paper's claim that the target orientation can be estimated without knowledge of the target shape is therefore not supported by the stated model.","section":"Section III-B, Eqs. (24) and (28)"},{"comment":"The translation estimator in Eq. (23) is described as a quadratic program, but the objective is a fourth-degree polynomial in t: \\hS is affine in t and the squared Frobenius norm turns the quadratic dependence into a quartic one. The text gives no feasible set, no initialization strategy, and no global-optimality argument; citing generic gradient descent or interior-point methods is insufficient for what is in general a nonconvex quartic problem. More importantly, since the \\hD2 contained in the objective comes from the invalid Eq. (14), the minimizer of Eq. (23) has no demonstrated relationship to the true translation t. I did not reproduce the separate claim that the objective is invariant in t, but the invalid completion is already decisive against the translation-estimation claim as presented.","section":"Section III-A, Eq. (23)"}],"minor_comments":[{"comment":"The notation t ⊗ 1_{N1}^⊤ in Eq. (19) is undefined; the intended object appears to be the outer product t 1_{N1}^⊤, not a Kronecker product.","section":"Section III-A, Eq. (19)"},{"comment":"The optimization over Q in Eq. (28) must enforce the orthogonality constraint Q^⊤ Q = I, but the text does not state how this constraint is handled. The paper mentions CVX, yet Eq. (23) is not convex and Eq. (28) is not a standard convex program; implementation details are needed.","section":"Section III-B, Eq. (28)"},{"comment":"The genie-aided baseline is not a clean upper bound for the effect of the Nyström step, because it still uses the completed \\hD2 from Eq. (14) in the MDS and translation-estimation pipeline; only Q is supplied externally. A genie that supplies the true D2 or C2 would be needed to isolate the invalidity of Eq. (14).","section":"Section IV, Figures 3 and 4"},{"comment":"The MDS embedding in Eq. (16) does not specify how negative eigenvalues of ̅D are handled under noise. Since ̅D may be indefinite, V Λ^{1/2} is not always real; a truncation or projection rule should be stated.","section":"Section III-A, Eq. (16)"},{"comment":"The matrices C1 and C2 contain typographical formatting errors (e.g., \"1 .5\", \"0 .5\") that should be corrected in a revision.","section":"Table I"}],"recommendation":"reject","confidential_remarks":"The manuscript has a clear problem statement and a reasonable simulation framework, but the core mathematical claims are not defensible. The Nyström completion in Eq. (14) is a technical error that corrupts the entire pipeline, and the rotation estimate is fundamentally unidentifiable without target-shape knowledge. These are not modest local issues; they affect the central contribution. I recommend rejection, though I would encourage the authors to consider reformulating the problem so that only the target point set S2 is estimated from cross-body distances, with orientation treated as an additional identifiable quantity only under extra priors."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know about this one. First, the problem it targets is real and, as far as the citations go, unaddressed: egoistic rigid body localization with unknown target shape and N1 ≠ N2. That alone is worth a careful look. Second, the central completion step, Eq. (14), is mathematically invalid. It applies Nyström to the raw distance matrix D, treating it like a PSD kernel, but an EDM is conditionally negative definite. The completion should act on the double-centered Gram matrix K = -1/2 J D^2 J; Eq. (14) is not equivalent. The stress-test note gives a clean 1D noiseless counterexample: the true D2 entry is 4, Eq. (14) gives 239.5. Once D2 is wrong, the MDS reconstruction of the target and everything downstream (translation and rotation) is built on garbage. So the simulations in Section IV, which use this pipeline, do not validate the method.\n\nThe reader also flagged Eq. (23) as invariant under t. I don't think that holds up—the stress-test note's counterexample is convincing, and a quick look at the quadratic form suggests the double-centering does not annihilate all t-dependence. But it is a side issue: Eq. (14) is enough to sink the paper.\n\nWhat the paper does well: it is clearly written, the problem statement is sharp, and the authors correctly identify a limitation of prior work (they even cite a flaw in [22]). The genie-aided comparison in Figure 3 is a reasonable idea, and the restriction to moderate-to-low ranging errors is honestly stated.\n\nThe soft spots beyond Eq. (14): footnote 8 conditions on rank(D1) ≥ rank(D2) without justification, and Eq. (28) for rotation relies on the same corrupted reconstruction. There is no proof for the Nyström step—just an \"yields.\"\n\nWho gets value from this: a researcher looking for a concrete failure mode in EDM-based localization, or someone who wants to fix the completion step and build on the egoistic problem. It is worth a serious referee because the problem statement is a real gap, and the fix might be straightforward. But as written, it is a reject. I would tell the authors to redo the completion using properly double-centered Gram matrices and re-run the simulations. If that works, the paper could become a useful contribution.","headline":"The egoistic setup is genuinely new, but the Nyström completion on a raw EDM is wrong, so the paper's pipeline collapses at step one.","tokens_in":10585,"tokens_out":3237,"would_cite":false,"duration_ms":31294,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A rigid body can estimate another body's relative translation and rotation using only cross-body distance measurements, without knowing the target's shape.","keywords":["rigid body localization","egoistic localization","multidimensional scaling","Nyström approximation","translation estimation","rotation estimation","autonomous driving","convex optimization"],"falsifier":"Take two rigid bodies with known shapes and choose N1 < N2, so the rank condition rank(D1) ≥ rank(D2) fails; measure noiseless cross-body distances and run the method. If the Nyström-completed D2 differs materially from the true D2, or if equations (23) and (28) produce t and Q far from the ground truth, the central claim fails.","tokens_in":9524,"feed_emoji":"🚗","tokens_out":4140,"duration_ms":40270,"temperature":0.7,"pith_summary":"This paper tries to establish that one rigid body, such as a truck, can estimate the relative translation and rotation of another rigid body, such as a car, using only measured distances between sensors on the two bodies, without knowing the target body's shape and even when the two bodies have different numbers of sensors. If true, this removes a major obstacle to using radio-based rigid-body localization in autonomous driving, because a vehicle no longer needs a pre-stored model of every possible target. The paper proposes a pipeline built on multidimensional scaling (MDS), in which the double-centering operator converts squared cross-body distances into inner products, a Nyström step completes the unknown intra-body distance matrix, and convex programs recover the translation vector and rotation matrix. Simulation results for a truck-car scenario with 12 and 10 landmarks show that the translation estimate stays close to a genie-aided version that knows the target shape.","feed_headline":"A car can read another car's pose from distances alone","feed_subtitle":"MDS-based scheme reconstructs the target's missing distances and solves for t and Q, even when car shapes and sensor counts differ.","key_machinery":"The central object is the Schönberg double-centering operator J = I - (1/N)11^T, which maps a squared distance matrix to a centered inner-product matrix and underlies equations (10), (17), and (18). Around it, the paper builds a chain: Nyström approximation in equation (14) completes the unknown intra-distance matrix D2 from the known D1 and measured D12; classical MDS in equation (16) reconstructs the combined sensor positions; a Procrustes alignment in equation (19) brings the reconstruction into body 1's frame; a quadratic program in equation (23) estimates the translation vector; and an eigendecomposition with permutation correction in equation (28) estimates the rotation matrix.","core_discovery":"The paper claims that the egoistic rigid-body localization problem—estimating the relative pose Q and t of a second body with unknown shape C2 from the cross-distance matrix D12—can be solved without knowing C2. The translation estimate is obtained by first reconstructing the target's intra-distance matrix D2 via Nyström completion, applying MDS to get a sensor-position estimate, aligning it to body 1's frame via Procrustes, and then solving the quadratic program in equation (23) for t. The rotation estimate is obtained from the double-centered cross-distance Gram matrix, whose product with its transpose equals QΛQ^T, and equation (28) searches over the six eigenvalue permutations of Λ to avoid eigenvector swaps.","pith_inferences":["The Nyström completion in equation (14) is the fragile link: if it fails, the whole pose estimate fails. Replacing it with a rank-constrained Euclidean distance matrix completion method would make the pipeline robust when the target has more landmarks than the ego body, an extension the paper does not explore.","The same egoistic geometry could be applied to non-vehicular rigid bodies, such as robot arms, drones, or handheld devices, whenever pairwise ranging is available, and to tracking by feeding each new frame's t and Q into a filter; both directions are left implicit.","The paper evaluates translation accuracy numerically but does not report a direct RMSE test of the rotation estimate against ground-truth yaw, pitch, and roll angles, including near-spherical bodies where eigenvector swaps occur; such a test would be the natural next experiment."],"forward_implications":["Autonomous vehicles can estimate the relative pose of nearby vehicles using only inter-vehicle range measurements, without requiring the target vehicle to broadcast its shape or sensor configuration.","The method works for bodies with different numbers of landmarks (N1 ≠ N2), so a vehicle with 12 sensors can localize a target with 10 sensors, a case that prior same-shape methods could not handle.","Translation estimation via equation (23) and rotation estimation via equation (28) are formulated as convex optimizations solvable by standard tools, making the approach computationally feasible for real-time V2X use.","In the low-to-moderate ranging error regime typical of automotive sensing, the egoistic translation estimate remains close to the genie-aided estimate that knows the target shape."],"supporting_citations":[{"why":"Introduces the anchorless multi-rigid-body localization problem and the Procrustes-based rotation estimation that this work adapts to the unknown-shape setting.","marker":"[22]"},{"why":"The state-of-the-art rigid-body localization method used as the baseline for translation RMSE comparison.","marker":"[16]"},{"why":"Classic MDS double-centering operator that maps squared distances to centered inner products, the core of equations (10), (17), and (18).","marker":"[25]"},{"why":"The generalized orthogonal Procrustes solution used in the alignment step of equation (19).","marker":"[26]"},{"why":"The Nyström approximation that equation (14) uses to complete the unknown intra-body distance matrix.","marker":"[30]"},{"why":"Supplies the squared-distance noise model and the anchor-based precursor of the tracking problem.","marker":"[23]"},{"why":"Journal version containing the proof that the translation step of [22] is incorrect, motivating the new translation estimator.","marker":"[24]"}],"fun_headline_variants":["MDS lets cars read each other's pose from distances alone","Anchorless vehicle pose from cross-distance measurements","Cars decode relative pose without knowing target shape","MDS-based method localizes rigid bodies egoistically"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole pipeline stands on the assumption that the Nyström completion in equation (14) recovers the target's unknown intra-body distance matrix from the measured cross-body distances and the known self-distances; if that completion is wrong, the reconstructed sensor positions, translation, and rotation are all wrong.","fun_headline_variants_meta":{"raw":{"variants":["MDS lets cars read each other's pose from distances alone","Anchorless vehicle pose from cross-distance measurements","Cars decode relative pose without knowing target shape","MDS-based method localizes rigid bodies egoistically"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000372,"raw_usage":{"total_tokens":1936,"prompt_tokens":836,"completion_tokens":1100,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":452,"completion_tokens_details":{"reasoning_tokens":1038}},"tokens_in":452,"tokens_out":1100,"duration_ms":8307,"temperature":1.0,"reasoning_tokens":1038,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T18:19:36.389844+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take two rigid bodies with known shapes and choose N1 < N2, so the rank condition rank(D1) ≥ rank(D2) fails; measure noiseless cross-body distances and run the method. If the Nyström-completed D2 differs materially from the true D2, or if equations (23) and (28) produce t and Q far from the ground truth, the central claim fails.","supporting_citations":[{"cited_title":"Towards multi-rigid body localization,","cited_arxiv_id":null,"evidence_quote":"Introduces the anchorless multi-rigid-body localization problem and the Procrustes-based rotation estimation that this work adapts to the unknown-shape setting."},{"cited_title":"Accurate localization of a rigid body using multiple sensors and landmarks,","cited_arxiv_id":null,"evidence_quote":"The state-of-the-art rigid-body localization method used as the baseline for translation RMSE comparison."},{"cited_title":"Multidimensional scaling: I. theory and method,","cited_arxiv_id":null,"evidence_quote":"Classic MDS double-centering operator that maps squared distances to centered inner products, the core of equations (10), (17), and (18)."},{"cited_title":"A generalized solution of the orthogonal procrustes problem,","cited_arxiv_id":null,"evidence_quote":"The generalized orthogonal Procrustes solution used in the alignment step of equation (19)."},{"cited_title":"Using the Nystr ¨om method to speed up kernel machines,","cited_arxiv_id":null,"evidence_quote":"The Nyström approximation that equation (14) uses to complete the unknown intra-body distance matrix."},{"cited_title":"Tracking position and orientation of a mobile rigid body,","cited_arxiv_id":null,"evidence_quote":"Supplies the squared-distance noise model and the anchor-based precursor of the tracking problem."}],"review_version":1}