{"id":"027be18f-6cc8-4808-9c56-34d7199c71ae","arxiv_id":"2508.01894","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":5.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"IMUCoCo maps IMU signals from any body-surface location into a coordinate-based feature space so that wearable sensors can be placed flexibly while still supporting pose estimation and activity recognition.","lead":"A new framework, IMUCoCo, translates signals from body-worn inertial sensors into a shared feature space that depends on where on the body each sensor sits. This could let people wear smart devices in any position they prefer while still getting accurate pose and activity tracking.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The submitted full text is not the IMUCoCo paper, and the abstract itself leaves coordinate acquisition unspecified, so the central flexibility claim is currently unverifiable.","rationale":"The reader's weakest assumption correctly identifies coordinate availability and generalization to unseen placements as the technical linchpin, and the abstract indeed says nothing about how coordinates are obtained. My read agrees with that. However, the dominant issue in the current review artifact is the full-text mismatch: the manuscript supplied under the arXiv ID is not the IMUCoCo paper, so the central claim cannot be assessed at all. This is not a flaw in the method itself but a load-bearing obstacle to verification, which is why the verdict should remain UNVERDICTED. I set agreement to 'partial' because the reader's stated weakest assumption is the coordinate/generalization issue, while I emphasize the missing methodology as the more immediate reason no technical judgment can be rendered. The concrete test proposed — checking whether held-out placements appear in the evaluation — would, if answered positively, substantially support the flexibility claim; if answered negatively or left unanswered, the claim remains unverified.","tokens_in":1019,"tokens_out":3138,"duration_ms":37257,"concrete_test":"Retrieve the actual arXiv:2508.01894 full text and check whether the test set for pose estimation includes body-surface placements that were never used in training and are not fine-tuned at inference; if the evaluation only covers placements present during training, the claimed flexibility to arbitrary placements is not supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The abstract's central claim is that mapping signals from a variable number of IMUs into a unified feature space based on their spatial coordinates enables accurate pose estimation at typical and atypical placements. This claim depends on two conditions that the submitted text does not establish: (i) each IMU's body-surface spatial coordinate is available at inference time, and (ii) the coordinate-conditioned mapping generalizes to placements and sensor counts not seen during training. The provided full text under arXiv:2508.01894 is actually an excursion-set cosmology manuscript, so neither condition can be checked against a methodology or evaluation. Even taking the abstract at face value, it says nothing about how coordinates are obtained — manual annotation, device self-localization, or learned estimation. If coordinates require manual measurement or user input at each placement, the claimed flexibility 'according to their needs and preferences' is weakened by an unstated user burden. If coordinates are estimated by another model, errors in that estimate propagate into the unified feature space. Without the actual method and a held-out placement evaluation, the core flexibility claim is unverified rather than demonstrated.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The submission presents IMUCoCo, a framework intended to map IMU signals from an arbitrary, variable number of body-surface sensors into a unified feature space conditioned on spatial coordinates, with downstream pose estimation and activity recognition. The abstract claims that the framework supports accurate pose estimation across typical and atypical sensor placements, enables users to change device locations by context, and suggests placements per use case. However, the supplied full text is not the IMUCoCo paper but an unrelated cosmology manuscript on primordial black holes in excursion set theory. Consequently, the submission contains no method description, no training or inference details, no evaluation protocol, no baselines, and no quantitative results for the claimed system. Every central claim in the abstract is therefore unverifiable from the submitted material.","tokens_in":1203,"tokens_out":2656,"duration_ms":31293,"significance":"If the claimed framework were actually implemented and validated, it would address a genuine and growing usability bottleneck in IMU-based motion sensing: freeing users from a small set of fixed, training-dictated sensor placements and allowing flexible, context-dependent placement. The problem statement is well motivated, and the idea of a continuous coordinate-conditioned feature space is a plausible and potentially valuable design direction. That said, the submission as received provides no evidence for the central claims and no artifact that could be checked: there are no machine-checked proofs, no reproducible code, no datasets, and no empirical results. The significance can only be judged as prospective, not demonstrated.","major_comments":[{"comment":"The full text under the submission is arXiv:2508.01896, 'Clustering of Primordial Black Holes in Excursion Set Theory', which is entirely unrelated to the IMUCoCo abstract. The manuscript therefore contains none of the claimed framework: no network architecture, no training objective, no coordinate encoding scheme, no downstream model, no evaluation protocol, and no results. This is a load-bearing failure because the central claim of flexible, accurate IMU-based pose estimation cannot be assessed in any way. This issue cannot be fixed by local revision; the actual IMUCoCo manuscript would have to be supplied in its entirety.","section":"Full Text (submitted manuscript)"},{"comment":"The sentence 'Our evaluations demonstrate that IMUCoCo supports accurate pose estimation in a wide range of typical and atypical sensor placements' is made without any quantitative support: no dataset is named, no baseline is compared, no error metric is reported, no ablation is described, and no error analysis is given. For a machine-learning systems paper, an empirical claim of state-of-the-art flexibility requires at least summary statistics and a clear held-out evaluation protocol; none appear in the submission.","section":"Abstract"},{"comment":"The framework's premise is that each IMU's spatial coordinate on the body surface is available at inference time, yet the abstract never states how these coordinates are obtained: by manual annotation, by device self-localization, by a learned estimator, or by some other mechanism. If a user must measure or enter coordinates at every placement, the promised flexibility is weakened by an unstated user burden; if coordinates come from another model, errors in that estimate propagate into the unified feature space. This condition is load-bearing for the central flexibility claim and is left unspecified.","section":"Abstract"},{"comment":"The claim that the model generalizes to 'a wide range of typical and atypical sensor placements' and to 'a variable number of IMUs' is not tied to any explicit training distribution or evaluation split. The abstract does not establish that the coordinate-conditioned mapping is tested on held-out placements and sensor counts rather than on placements that shaped the model during training, leaving the core flexibility claim unsubstantiated.","section":"Abstract"}],"minor_comments":[{"comment":"The title, abstract, and full text describe entirely different papers; even a reader who only skims the submission will be confused about the subject matter.","section":"General"},{"comment":"The submission contains no references, figures, or equations for the IMUCoCo work, making it impossible to follow or verify any part of the proposed method.","section":"General"}],"recommendation":"reject","confidential_remarks":"The editor may wish to verify that the correct full text was uploaded for this submission. As received, the article is not a coherent submission: the abstract describes an IMU-based pose-estimation framework, while the full text is a cosmology paper. Even setting aside that mismatch, the abstract alone does not provide enough methodological or empirical detail to support the central claims. A fresh review would be required if the correct manuscript is made available."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The submitted full text is not the IMUCoCo paper. It is a manuscript on primordial black hole clustering in excursion set theory. So I cannot review a method or results that are not in front of me. I'll say what I can about what I actually have: the abstract.\n\nThe abstract's core idea is genuinely interesting. Mapping IMU signals from arbitrary body locations into a coordinate-conditioned feature space is a sensible departure from the fixed-placement status quo. If the learned mapping actually generalizes to placements and sensor counts not seen during training, it would remove a real usability bottleneck in wearable sensing. That is worth something.\n\nBut the abstract alone leaves the two load-bearing questions open. First, how are the spatial coordinates of each IMU obtained at inference time? If manual annotation is required, the claimed flexibility is weakened by an unstated user burden. If coordinates come from another model, errors in that estimate propagate directly into the feature space. The abstract says nothing about this. Second, there is no evidence that the evaluation uses held-out placements. If the training set already covers the tested locations, the claim of flexibility reduces to interpolation, not generalization. The abstract also provides no numbers, baselines, ablations, or dataset descriptions, so I cannot even gauge the strength of the empirical claim.\n\nNone of this is a takedown of IMUCoCo as a research direction. The idea may well work. But the submission as it stands is missing its own paper. That is a serious procedural problem, not a minor flaw. The correct move for a desk editor is to verify the file, not to send this abstract-plus-cosmology-paper to reviewers.\n\nIf the actual IMUCoCo paper delivers on the abstract—with a credible coordinate-acquisition method, held-out placement testing, and comparison to fixed-placement baselines—then it would deserve a serious referee. On the evidence in front of me, I cannot recommend peer review for this submission.","headline":"The full text under this arXiv ID is a cosmology paper, so the only thing on the table is an abstract that promises a useful idea but leaves the key questions unanswered.","tokens_in":1701,"tokens_out":1289,"would_cite":false,"duration_ms":16853,"reading_group":"no","serious_thinker":"unclear","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"IMUCoCo claims that pairing each IMU signal with its body-surface coordinate and mapping these pairs into a unified feature space lets pose estimation and activity recognition work across a wide range of sensor placements, including…","keywords":["IMUCoCo","inertial measurement units","human pose estimation","activity recognition","sensor placement flexibility","continuous coordinates","wearable devices"],"falsifier":"Train an IMUCoCo model on wrist, chest, and ankle placements, then evaluate it with a single IMU placed on the crown of the head or the sole of the foot; if pose error is no better than a coordinate-blind baseline, the claimed flexibility is not supported.","tokens_in":850,"feed_emoji":"⌚","tokens_out":8978,"duration_ms":99861,"temperature":0.7,"pith_summary":"IMUCoCo claims that the standard practice of placing inertial measurement units at fixed, model-prescribed body locations is unnecessary. The framework takes a variable number of IMU signals from arbitrary points on the body surface, pairs each with its spatial coordinate, and maps these pairs into a shared feature space that downstream pose-estimation and activity-recognition models can use. On that basis the authors report accurate full-body pose estimation across both typical and atypical sensor placements. If the claim holds, everyday devices equipped with IMUs could be worn wherever is comfortable, moved between contexts, and still support motion sensing, with the system even suggesting where to place sensors for a given use case.","feed_headline":"IMUCoCo frees body-worn sensors from fixed placement spots","feed_subtitle":"Coordinate-conditioned mapping lets a model handle sensors at typical and atypical body locations without retraining.","key_machinery":"The load-bearing mechanism is the continuous-coordinate conditioning step: each IMU's inertial signal is combined with the IMU's body-surface coordinate and encoded into a unified feature space, so downstream models see placement-aware features rather than signals tied to named body sites. Because the conditioning variable is continuous, the same encoder can in principle accept any number of sensors at any location on the body surface. The paper names this framework IMU over Continuous Coordinates (IMUCoCo).","core_discovery":"The central claim is that sensor location can be treated as a continuous input rather than a discrete categorical slot. By conditioning each IMU's signal on its body-surface coordinate and projecting the result into a unified feature space, IMUCoCo aims to make the learned representation independent of any specific placement configuration. The paper reports that this representation supports accurate pose estimation across a wide range of typical and atypical placements, works with a variable number of sensors, and also enables use-case-dependent placement suggestions.","pith_inferences":["The supplied full text is a different manuscript about primordial black hole clustering, so the abstract and metadata are the only evidence for IMUCoCo's empirical claims in this record.","If the coordinate-to-feature mapping is continuous over the body surface, features for unmeasured locations could be interpolated, in effect creating virtual sensors where no device is worn; the paper does not claim this directly.","A practical deployment likely needs a way to obtain each IMU's body-surface coordinate automatically, for example by self-localization from the IMU signals, because manual measurement would undermine the flexibility the framework promises.","Training on heterogeneous placements from many users could enlarge the available motion data, but cross-user use may require normalizing coordinates across different body shapes."],"forward_implications":["A user could move an IMU-equipped device from wrist to ankle, pocket, or elsewhere between sessions or contexts without retraining, provided the new coordinate is supplied.","Because the feature space is unified, the same downstream pose and activity models can accept different numbers of IMUs, so adding or removing a sensor does not require a new model.","The placement-suggestion capability could tell users where to put sensors to maximize accuracy for a specific activity or use case.","If the mapping generalizes smoothly, sparse training placements could support many untrained body locations rather than only the handful of predefined spots used today.","Both pose estimation and activity recognition can be plugged into the same IMUCoCo feature space, so one sensing framework serves both tasks."],"supporting_citations":[],"fun_headline_variants":["IMUCoCo: any body placement works for pose and activity sensing","Treat IMU location as a continuous input: IMUCoCo","Forget fixed sensor spots: IMUCoCo adapts to any placement","Flexible IMU placement: coordinates, not categories","IMUCoCo lets you place sensors anywhere, still tracks pose"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The framework assumes that each IMU's body-surface coordinate is known or inferable at inference time and that the coordinate-conditioned mapping learned from training placements generalizes to body locations and sensor counts never seen in training.","fun_headline_variants_meta":{"raw":{"variants":["IMUCoCo: any body placement works for pose and activity sensing","Treat IMU location as a continuous input: IMUCoCo","Forget fixed sensor spots: IMUCoCo adapts to any placement","Flexible IMU placement: coordinates, not categories","IMUCoCo lets you place sensors anywhere, still tracks pose"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000919,"raw_usage":{"total_tokens":3905,"prompt_tokens":872,"completion_tokens":3033,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":488,"completion_tokens_details":{"reasoning_tokens":2942}},"tokens_in":488,"tokens_out":3033,"duration_ms":23128,"temperature":1.0,"reasoning_tokens":2942,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T05:18:03.868447+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Train an IMUCoCo model on wrist, chest, and ankle placements, then evaluate it with a single IMU placed on the crown of the head or the sole of the foot; if pose error is no better than a coordinate-blind baseline, the claimed flexibility is not supported.","supporting_citations":[],"review_version":1}