{"id":"71b4ca4f-fa85-4221-b235-831b46b95fd1","arxiv_id":"2505.24348","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"This paper proposes a crowdsensing framework that combines gamified AR and passive wearable LiDAR collection, Geohash data management, and point cloud registration to keep urban digital twins updated.","lead":"This paper presents a mobile crowdsensing framework that uses LiDAR-equipped phones, an AR territory game, and a wearable neck sensor to collect 3D point clouds and merge them into a city-scale digital twin. A reader might care because it is a concrete attempt to keep urban digital twins updated with citizen data instead of dedicated survey fleets.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Registration claim rests on masking a clean city-scale cloud; real 3D-MCS scans contain noise, occlusion, and drift, so the <40% removal-ratio result may not transfer to actual deployments.","rationale":"The reader's weakest-assumption analysis identified exactly the same load-bearing concern: the registration evaluation uses a randomly masked subregion of a clean, publicly available city model as a proxy for real crowdsourced point clouds, and real MCS captures are not clean rigid subsets of the same model. I agree with this diagnosis and with the CONDITIONAL verdict. My stress-test pass did not find an internal inconsistency that would justify rejection; the framework is implemented, the experiments are described transparently, and the limitations are partially acknowledged in the conclusion. However, the central effectiveness claim for the dynamic integration mechanism rests on an experiment that removes the very noise, partial overlap, and sensor artifacts that define the problem. A concrete follow-up using the authors' own collected point clouds against a reference map would settle whether the reported 40% threshold transfers to actual deployments. Because the paper explicitly frames the integration experiment as an assumption ('we assume that partial point clouds collected by 3D-MCS'), the current claim is best understood as a synthetic feasibility study rather than verified real-world performance. I would keep the verdict CONDITIONAL, with the condition being real-data registration validation.","tokens_in":9072,"tokens_out":3326,"duration_ms":44828,"concrete_test":"Run the Section III-F pipeline on real point clouds collected by the paper's own active/passive apps (e.g., the campus walk in Section III-C or the filtered data in Section III-D) and attempt to register them against a high-quality reference map of the same area, such as VIRTUAL SHIZUOKA or a survey-grade LiDAR scan. Compute registration success rate, RMSE, and processing time for the same V and N settings as Figures 16–18, both with and without the paper's noise-removal and density-equalization preprocessing. If the success rate or RMSE degrades materially relative to the masked-subset results, the '<40% removal ratio' claim must be restated as a property of clean synthetic subsets, not of real 3D-MCS data.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's strongest quantitative claim is the Section III-F conclusion that the dynamic point cloud integration mechanism 'can successfully register point clouds with low RMSE and short processing time when the removal ratio is approximately lower than 40%.' This is supported only by Figure 15's synthetic experiment: a 12.3 m × 15.2 m subregion of VIRTUAL SHIZUOKA is extracted and its center is masked with a rectangle, then the masked cloud is registered back against the original region. The query and target therefore share the same point density, noise characteristics, coordinate frame, and surface appearance; the only difference is the deliberately removed central patch. Real 3D-MCS captures—including those shown in Sections III-A and III-C—contain sensor noise, moving objects, incomplete surfaces, viewpoint-dependent occlusion, and ARKit/self-localization drift. These are exactly the conditions the framework must handle, yet the experiment explicitly omits preprocessing 'since we do not add additional noises in the dataset.' Consequently, the reported success rates, RMSE, and processing times may measure recovery of a known rigid transform on clean data rather than registration of genuine crowdsourced scans. The abstract's broader claim that effectiveness was 'verified' from real-world experiments is therefore not supported for the integration mechanism; the real-world component is limited to data collection demonstrations, while the registration analysis is synthetic. This is the load-bearing weakness because the dynamic integration mechanism is the core technical contribution that makes the framework more than a collection of apps.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes a 3D mobile crowdsensing framework for sustainable urban digital twins, comprising an active AR gamified sensing model, a passive wearable sensing model, Geohash-based spatial management, a dynamic point cloud integration mechanism based on global and local registration, and a web-based real-time visualizer. The authors report real-world field experiments on a university campus, a 30-subject questionnaire study on gamification, point cloud filtering and data-size analyses, and a registration experiment using the VIRTUAL SHIZUOKA dataset. The paper claims that the framework's effectiveness is verified by these experiments and that the dynamic integration mechanism can register point clouds with low RMSE and short processing time when the removal ratio is below about 40%.","tokens_in":9336,"tokens_out":5492,"duration_ms":76888,"significance":"The paper addresses a genuine gap: keeping urban digital twins current with spatial data collected by ordinary LiDAR-equipped mobile devices. Its strengths include a working end-to-end implementation with real field collection in both active and passive modes, a public city-scale dataset for the registration study, and useful measurements of data-size and processing-time breakdowns. If the registration results transferred to actual crowdsourced point clouds, the stated removal-ratio guideline would be a practical design rule. However, the evidence base is currently narrow: the registration evaluation is a synthetic self-consistency test, the subjective evaluation is small and descriptive, and some conclusions go beyond what the experiments can support.","major_comments":[{"comment":"The central quantitative claim that the dynamic integration mechanism 'can successfully register point clouds with low RMSE and short processing time when the removal ratio is approximately lower than 40%' rests on a test in which a clean rectangular subregion of VIRTUAL SHIZUOKA is masked and registered back against the original cloud. The query and target share the same point density, noise characteristics, coordinate frame, and surface appearance, so the experiment measures recovery of a known rigid transform on clean data. Real 3D-MCS captures—including those shown in Sections III-A and III-C—contain sensor noise, moving objects, viewpoint-dependent occlusion, and ARKit self-localization drift; the authors explicitly skip preprocessing because no noise is added. This means the reported success rate, RMSE, and processing time do not yet establish the abstract's and conclusion's claims that effectiveness was 'verified' in real-world settings. I recommend adding integration experiments with actual MCS scans against a UDT, or at least a controlled injection of realistic noise, occlusion, and pose error with sensitivity analysis.","section":"Section III-F, Figures 15–18"},{"comment":"The success-rate methodology is under-specified. The text describes one randomly selected 12.3 m × 15.2 m subregion, but Figure 16 reports success rates without stating the number of random subregions or independent trials per condition, any confidence intervals, or the random seed. If each success rate is based on a single random draw per parameter setting, the percentage is not statistically meaningful. Please report the number of independent trials and the variance across trials; a baseline comparison with an alternative registration pipeline would also strengthen the claim.","section":"Section III-F, Figure 16"},{"comment":"The subjective evaluation does not support the strong claim that gamification 'was effective in enhancing motivation.' The study uses 30 self-selected students, one minute of exposure per condition, no reported randomization or counterbalancing, and no inferential statistics. The standard deviations for Q1–Q4 are large relative to the mean differences, and Q5–Q6 have no non-gamified comparator by construction, so they cannot substantiate behavioral change. Please report significance tests or effect sizes with appropriate multiple-comparison correction, or revise the conclusion to describe the results as preliminary evidence of user perception.","section":"Section III-B, Table I"},{"comment":"The manuscript itself states that the current implementation does not yet address the quality of the collected point cloud data or the potential benefits of re-collecting data in previously scanned areas. That limitation bears directly on the framework's core value proposition of updating a UDT with newly collected crowdsourced data. Given this caveat, the conclusion's phrase 'comprehensively evaluated' overstates the evidence. I recommend either adding an experiment that measures the benefit or risk of updating previously captured regions, or substantially tempering the comprehensiveness claim.","section":"Section IV, final paragraph"}],"minor_comments":[{"comment":"There are several typographical errors: 'locaiton' in Figure 8, 'crowdsesning' in the Index Terms, 'patial' in Section III-F, and 'integraiton' in Figure 15. Please correct these.","section":"Throughout"},{"comment":"The parameters N and V appear in Figure 16 and the surrounding text but are not defined in this section; N should be explicitly identified as the RANSAC parameter and V as the voxel size. The FPFH parameters are stated once but not defined symbolically.","section":"Section III-F"},{"comment":"The claim that the filtering process 'effectively removes' unreliable points is supported only by visual comparison in Figure 12; adding a quantitative metric, such as the fraction of points removed that are later judged inconsistent, would make the evaluation more convincing.","section":"Section III-D"},{"comment":"The observation that AR map data size converges to a fixed value regardless of area size is interesting but is reported without quantitative convergence values or a plot; please provide the actual numbers or a figure supporting this claim.","section":"Section III-E"},{"comment":"Reference [11] has a typo in the journal name ('IEEE Commun. Surveys tuts.'), and there are minor formatting issues such as 'V ol.' in reference [4]. Please proofread the reference list.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The paper is best read as a systems-integration paper whose components have appeared in earlier conference publications by the same group (references [7]–[10]). The new contribution here is the overall framework plus the evaluation, so the evaluation quality is decisive. I found no indication of fabrication or p-hacking; the concern is that the current experiments are too narrow to support the paper's strongest claims. The registration experiment in particular needs to be redone or substantially augmented before publication. The topic is within the scope of the journal, and the implementation appears to work as demonstrated."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The useful thing here is the integration itself. The authors took four mechanisms they had prototyped separately - gamified active sensing, passive wearable sensing, Geohash-based data management, and RANSAC+ICP registration - and actually built them into one working framework with iOS clients, a Node.js server, and a Three.js visualizer. The field demonstrations are real: point clouds were collected on campus, filtered, visualized, and merged, and the subjective comparison of gamified versus non-gamified sensing, while small, is a legitimate usability check. The Geohash data-size experiment is also a reasonable engineering measurement. The authors deserve credit for shipping a system, not just a proposal, and for explicitly noting in the conclusion that data quality and re-collection benefits are not yet addressed.\n\nThe soft spot is the quantitative centerpiece. The claim that registration works when the removal ratio is below about 40% comes from masking a rectangular region of a clean VIRTUAL SHIZUOKA point cloud and registering it against the original. That is a self-consistency test: same noise characteristics, same density, same coordinate frame, only a hole removed. The paper even says preprocessing was skipped because no noise was added. Real 3D-MCS captures, including the ones shown in Section III, contain drift, moving objects, lighting artifacts, and unreliable points. So the reported success rates and RMSEs are probably optimistic, and the abstract's phrase 'verified' through real-world experiments does not quite cover the integration mechanism, which was only 'analyzed' on a dataset. That distinction is in the abstract, but the conclusion and the Section III-F text make the 40% threshold sound more general than the evidence supports.\n\nThe other issues are minor in comparison. The subjective evaluation is 30 students for one minute, with no significance tests; I would treat it as suggestive, not confirmatory. The components are extensions of the authors' own papers, so the novelty is the integration and evaluation, not any single algorithm, which is a fair but modest claim. No code or data is released, which makes the registration results hard to check independently.\n\nOverall, I think the paper is a reasonable systems contribution for the smart-city and mobile-crowdsensing community. It deserves a serious referee, but the referee should push for either a real noisy MCS registration experiment or a much more hedged statement of the 40% threshold. I would not desk-reject it; I would send it out with a request for revision.\n\nRecommendation: send to peer review, and tell the authors to redo or reframe the registration evaluation.","headline":"A genuinely implemented integration of four MCS components, but the registration headline rests on a clean synthetic subset test and the real-world evidence is thinner than the abstract suggests.","tokens_in":9910,"tokens_out":1503,"would_cite":true,"duration_ms":23469,"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 mobile crowdsensing framework can keep urban digital twins current by merging game-driven and wearable LiDAR scans into a city point cloud, with accurate alignment demonstrated when the missing region is below about 40 percent.","keywords":["urban digital twin","mobile crowdsensing","LiDAR point cloud","gamification","augmented reality","Geohash","point cloud registration","smart city"],"falsifier":"Run the same RANSAC-plus-ICP pipeline on real captures from a LiDAR phone walking through an area already present in an urban point cloud, and compare success rate, RMSE, and processing time with the reported 20 and 40 percent removal results. If success drops materially under realistic noise, moving objects, or depth and confidence outliers, the integration claim does not transfer from the masked dataset to deployment.","tokens_in":8877,"feed_emoji":"🏙️","tokens_out":7860,"duration_ms":88723,"temperature":0.7,"pith_summary":"This paper proposes a complete mobile crowdsensing framework for keeping urban digital twins current using LiDAR-equipped phones carried by ordinary city residents. It combines two collection modes: an active augmented-reality game in which players color territories by scanning their surroundings, and a passive wearable mode in which a phone hung around the neck records the wearer's daily route. A Geohash layer indexes uploaded scans by location, and a server-side pipeline registers each scan into the city-scale point cloud using RANSAC global alignment followed by iterative closest point refinement. In the reported experiments, gamification raised positive affect and lowered negative affect among 30 users, and a masking study on a city-scale point cloud found registration success with low RMSE and short processing time when the removed region stayed below roughly 40 percent. The contribution is the framework design plus component-level evidence, not a long-term deployment.","feed_headline":"Phone scans refresh city twins while gaps stay under 40%","feed_subtitle":"Volunteers' game and wearable LiDAR scans merge into city digital twin clouds with low error and short run time.","key_machinery":"The load-bearing mechanism is the dynamic point cloud integration pipeline. It preprocesses each incoming scan with statistical outlier removal and voxel-grid density equalization, computes FPFH feature descriptors, matches them against a partial region of the urban digital twin, applies RANSAC to estimate the global transform, and refines the alignment with iterative closest point. This pipeline is what converts raw unaligned phone captures into registered updates, and the reported success rate, RMSE, and processing time all measure its behavior. The Geohash management layer feeds the pipeline by keeping scans and game information organized per region.","core_discovery":"The central claim is that sustainable urban digital twins can be fed by crowdsourced 3D scans from LiDAR-enabled mobile devices, provided the system solves incentives, storage, and alignment. The paper argues that its four-mechanism framework does this: the AR territory coloring game motivates players to cover both new and previously colored areas; the passive neck-worn model collects data during ordinary walking; Geohash keeps game state and point clouds indexed by region so the system can scale; and the dynamic integration pipeline merges partial scans into the existing twin. The quantitative anchor is the registration experiment: subregions of a city point cloud were masked at increasing removal ratios and re-registered, and the pipeline achieved high success rate and low root-mean-square error at 20 and 40 percent removal, with feature matching dominating processing time. The paper therefore concludes that dynamic integration works when the missing part of a captured region is approximately below 40 percent.","pith_inferences":["A direct test with real phone LiDAR captures, rather than clean masked subregions of the same point cloud, would likely shift the 40 percent threshold; the masking experiment models missing data but not sensor noise, moving objects, or trajectory drift.","Because the AR map information converged to a fixed size regardless of area size, larger game regions may lose reconstruction fidelity; a longer, larger-area deployment could test whether the game still guides participants to all needed spots.","The framework could prioritize re-scanning stale or incomplete regions using age-of-information scheduling, which the conclusion names as future work.","The web visualizer opens a human-in-the-loop path: when automatic RANSAC alignment fails, a user could nudge the cloud into place, effectively lowering the minimum overlap the system requires."],"forward_implications":["If the 40 percent registration threshold holds in practice, city-scale digital twins can be refreshed by ordinary LiDAR-phone carriers without survey-grade equipment, as long as each new scan overlaps the existing twin enough.","The subjective evaluation implies that gamification can strengthen motivation and steer participants toward uncolored as well as previously colored areas, addressing the incentive problem in mobile crowdsensing.","The passive wearable model implies that spatial data can be harvested along everyday walking routes without interrupting the carrier, expanding coverage to roads and paths.","The Geohash-based mechanism implies that the framework can be distributed across multiple servers by region and that registration searches can be limited to nearby regions.","Filtering by depth and confidence level removes unreliable points and cuts data size substantially, which lowers upload and storage costs."],"supporting_citations":[{"why":"Earlier version of the gamified sensing concept; the active model here extends this game design.","marker":"[7]"},{"why":"Earlier point-cloud collection experiment behind the passive wearable model.","marker":"[8]"},{"why":"Earlier study of dynamic point cloud integration that the registration pipeline builds on.","marker":"[9]"},{"why":"Earlier real-time visualizer concept used to display the merged passive-model point clouds.","marker":"[10]"},{"why":"Game experience questionnaire whose core module supplies the subjective motivation items Q1-Q4.","marker":"[11]"},{"why":"Public city-scale point cloud dataset used to evaluate registration success rate, RMSE, and processing time.","marker":"[12]"}],"fun_headline_variants":["Crowdsourced phone scans keep city twins fresh up to 40% gap","AR game and passive neck scans feed urban digital twins","Gamified crowdsensing merges point clouds into city twins","City twin updates from crowdsourced 3D scans with 40% missing","Phone-based crowdsensing fills city twin gaps under 40% loss"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The registration numbers assume that a mobile crowdsourcing capture resembles a clean, rigid subregion carved out of the same city-scale point cloud; real phone LiDAR scans contain sensor noise, moving objects, lighting artifacts, and trajectory drift, and the masking experiment removes none of those.","fun_headline_variants_meta":{"raw":{"variants":["Crowdsourced phone scans keep city twins fresh up to 40% gap","AR game and passive neck scans feed urban digital twins","Gamified crowdsensing merges point clouds into city twins","City twin updates from crowdsourced 3D scans with 40% missing","Phone-based crowdsensing fills city twin gaps under 40% loss"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000287,"raw_usage":{"total_tokens":1694,"prompt_tokens":962,"completion_tokens":732,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":578,"completion_tokens_details":{"reasoning_tokens":640}},"tokens_in":578,"tokens_out":732,"duration_ms":9596,"temperature":1.0,"reasoning_tokens":640,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T12:24:32.277307+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same RANSAC-plus-ICP pipeline on real captures from a LiDAR phone walking through an area already present in an urban point cloud, and compare success rate, RMSE, and processing time with the reported 20 and 40 percent removal results. If success drops materially under realistic noise, moving objects, or depth and confidence outliers, the integration claim does not transfer from the masked dataset to deployment.","supporting_citations":[{"cited_title":"Preliminary study of gamified 3D mobile sensing using LiDAR-equipped mo- bile devices,","cited_arxiv_id":null,"evidence_quote":"Earlier version of the gamified sensing concept; the active model here extends this game design."},{"cited_title":"The game experience questionnaire,","cited_arxiv_id":null,"evidence_quote":"Game experience questionnaire whose core module supplies the subjective motivation items Q1-Q4."},{"cited_title":"[Online]","cited_arxiv_id":null,"evidence_quote":"Public city-scale point cloud dataset used to evaluate registration success rate, RMSE, and processing time."}],"review_version":1}