{"id":"2000b2bc-42b2-400e-8e08-b05fbd467c12","arxiv_id":"2508.11485","paper_version":3,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":6.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":2,"one_line_summary":"i2Nav-Robot is a publicly released 17-km indoor-outdoor multi-sensor navigation dataset with centimeter-level reference trajectories from post-processed GNSS/INS, covering streets and parking lots.","lead":"This paper describes a new public robot navigation dataset, about 17 kilometers of indoor and outdoor driving with LiDARs, 4D radar, cameras, IMU, GNSS, and wheel odometry, plus centimeter-level reference positions. A generalist should care because such synchronized multi-sensor data with reliable ground truth is the testbed that autonomous vehicle navigation research needs.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Indoor GNSS-denied segments lack independent ground-truth validation; the 'fully covered centimeter-level' claim rests on unaudited IMU dead-reckoning.","rationale":"The reader's weakest assumption and my independent review converge on the same point: the ground-truth trajectory in GNSS-denied indoor sections is the most fragile element of the central claim. The abstract promises centimeter-level accuracy over the full route, but the generation method described (post-processed integrated navigation with a high-grade IMU) is exactly the kind of solution that needs external anchors to sustain that accuracy indoors. No such anchors are mentioned. The manuscript-body mismatch further prevents checking the actual processing pipeline, so the dataset cannot be verified from the paper alone. I agree with the reader's UNVERDICTED verdict: the dataset may be excellent, but this submission provides insufficient evidence. My concrete test, if run on the repository data, would settle whether the indoor ground-truth concern actually lands; until then, no change to the verdict is warranted.","tokens_in":16603,"tokens_out":2457,"duration_ms":32660,"concrete_test":"Download the dataset's post-processed trajectories and raw GNSS/LOAM/odometry logs from the GitHub repository. For each sequence that traverses indoors, identify transitions where the vehicle re-enters GNSS coverage. Compare the dead-reckoned indoor trajectory endpoint (just before reacquisition) with the GNSS-fixed position immediately after reacquisition; the discontinuity is an upper bound on accumulated indoor drift. If any such discontinuity exceeds 0.1 m, the 'centimeter-level' claim is falsified. Additionally, check whether the repository provides any surveyed control points or loop-closure residuals inside the parking areas; if none exist, note that the claim rests on unverified self-consistency.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The abstract's central claim is that i2Nav-Robot provides high-rate, reliable, fully covered, centimeter-level ground truth across ~17 km, including indoor parking lots. The stated derivation is post-processed integrated navigation using a high-grade IMU. Outdoors this can be constrained by GNSS, but indoors (parking garages, covered areas) the solution must rely primarily on IMU and wheel odometry. The abstract provides no independent verification for those segments—no surveyed control points, no loop-closure constraints, no reported per-segment accuracy. On an omnidirectional wheeled vehicle, wheel-slip and IMU-drift can degrade dead-reckoning beyond the claimed centimeter level over long indoor traverses. If the indoor reference drifts even modestly, the 'fully covered centimeter-level' claim fails precisely in the regime the dataset is designed to support, and the reported evaluation of 15 multi-sensor fusion methods against that reference becomes unreliable. Additionally, the supplied manuscript body is a different paper (VeteranAD, arXiv:2508.11488), so no methodological detail about the ground-truth generation is available in this submission to check the claim. The load-bearing weakness is therefore the unvalidated indoor ground-truth accuracy.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The submission arXiv:2508.11485 claims to introduce i2Nav-Robot, a large-scale indoor-outdoor multi-sensor fusion robot dataset. The abstract states that the dataset comprises ten sequences totaling about 17,060 meters, with multi-modal sensors (solid-state LiDARs, 4D MMW radar, stereo cameras, IMU, GNSS, wheel odometry), accurate timestamps from hardware synchronization and offline calibration, centimeter-level ground truth from post-processed integrated navigation using a high-grade IMU, and an evaluation using 15 open-source multi-sensor fusion navigation methods. However, the full text of the manuscript is a completely different paper on end-to-end autonomous driving (VeteranAD, arXiv:2508.11488), with no description of the i2Nav-Robot platform, sensor suite, calibration, synchronization, ground-truth generation, or evaluation. Consequently, the manuscript as submitted does not contain the substance of the claimed dataset paper and its central claims cannot be assessed.","tokens_in":16929,"tokens_out":2449,"duration_ms":30198,"significance":"If the i2Nav-Robot dataset exists as described, it could be a valuable contribution to multi-sensor fusion navigation research: it combines modern solid-state LiDARs, 4D MMW radar, stereo vision, IMU, GNSS, and wheel odometry, and it targets the important but under-served regime of indoor-outdoor transitions and GNSS-denied indoor environments. A publicly available dataset with synchronized multi-modal data and independent centimeter-level reference trajectories would be useful for benchmarking fusion algorithms. However, the submitted manuscript provides no concrete evidence for these claims: no sensor specifications, no calibration or synchronization validation, no ground-truth accuracy analysis, no sequence table, and no evaluation results. The only verifiable asset is a GitHub link. The paper therefore currently offers no basis for the claimed significance.","major_comments":[{"comment":"The full text of the manuscript is not about i2Nav-Robot. It is a paper titled \"Perception in Plan: Coupled Perception and Planning for End-to-End Autonomous Driving\" (VeteranAD, arXiv:2508.11488), by different authors and with different subject matter. None of the dataset description, sensor configurations, calibration procedures, synchronization details, ground-truth generation, or evaluation of 15 multi-sensor fusion methods mentioned in the abstract appears in the body. This is a load-bearing mismatch: the central claim of the paper cannot be checked because the actual submission does not contain the claimed dataset paper.","section":"Full text / Abstract"},{"comment":"The abstract asserts \"high-rate, reliable, and fully covered ground truth, with centimeter-level positioning\" derived from post-processed integrated navigation using a high-grade IMU. No independent validation is provided for the GNSS-denied segments, such as indoor parking lots, where the reference trajectory depends primarily on IMU and wheel-odometry dead reckoning. Without surveyed control points, loop-closure constraints, or per-segment error statistics, the claim that centimeter-level accuracy holds throughout the full 17 km route is unsupported. Since the dataset is explicitly intended for indoor-outdoor navigation, this is not a peripheral issue: an inaccurate indoor reference would undermine the dataset's primary utility.","section":"Abstract, ground-truth claim"},{"comment":"The abstract states that \"accurate timestamps are obtained through both online hardware synchronization and offline calibration\" for all sensors. No timing-error statistics, synchronization validation, or calibration residuals are reported anywhere in the manuscript. For a multi-sensor fusion dataset, synchronization accuracy is a central quality metric; the claim is therefore unverifiable in the submitted text.","section":"Abstract, time synchronization claim"},{"comment":"The abstract claims that the dataset is \"evaluated by 15 open-sourced multi-sensor fusion navigation methods, demonstrating its superior data quality and utility.\" No experimental protocol, metric definitions, baseline names, result tables, or comparisons appear in the manuscript. The evaluation claim is load-bearing for the paper's stated contribution, and it is entirely absent.","section":"Abstract, evaluation claim"}],"minor_comments":[{"comment":"The title and abstract refer to i2Nav-Robot, but the body is a different paper on VeteranAD. The manuscript appears to be a file mix-up; this should be corrected at the submission level.","section":"Title vs. content"},{"comment":"Even if the correct body were supplied, a dataset paper would need a table of sequences with lengths, durations, environments, sensor specifications, and coordinate frames. None of this is present in the submitted text.","section":"Dataset contents"},{"comment":"The abstract references a GitHub repository, but the manuscript provides no license, download instructions, file format descriptions, or data usage guidelines.","section":"Availability and documentation"},{"comment":"The manuscript does not position i2Nav-Robot against existing multi-sensor fusion datasets (e.g., KITTI, nuScenes, UrbanNav, or similar robot datasets), so the claimed novelty and gap cannot be evaluated.","section":"Related work"}],"recommendation":"reject","confidential_remarks":"The submission is a different paper from the one described in its title and abstract. This is not a fixable technical issue within the manuscript as submitted; the central contribution is missing. I would suggest the editor return the submission so that the correct manuscript can be uploaded. I also note the abstract cites only a GitHub link and provides no independent validation of the ground-truth accuracy; if the corrected manuscript appears, those points should receive close scrutiny."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe headline is that the body of this arXiv submission is not the i2Nav-Robot dataset paper. It is VeteranAD (arXiv:2508.11488), an end-to-end driving paper. So we can only assess the abstract, which makes a real but unverified claim.\n\nWhat looks genuinely useful: the sensor suite is unusual in public UGV datasets—front and 360-degree solid-state LiDARs plus 4D MMW radar, stereo, IMU, GNSS, wheel odometry—and the 17 km indoor-outdoor coverage with ten sequences is a sensible response to a real gap. The plan to get accurate timestamps by combining hardware synchronization with offline calibration is the right approach. Using post-processed integrated navigation with a high-grade IMU to make reference trajectories is a legitimate methodology, and the group behind it has the relevant expertise.\n\nWhere it softens: the abstract's load-bearing sentence is 'high-rate, reliable, and fully covered ground truth, with centimeter-level positioning.' Indoors, where GNSS is gone, that claim rests entirely on IMU/wheel-odometry dead reckoning (with possibly other sensors). There are no surveyed control points, no loop closures, no per-segment accuracy numbers, no timing-error statistics in the abstract. Without those, 'fully covered centimeter-level' is an assertion, not a demonstrated property. The evaluation with 15 open-source methods is a good sanity check, but since the body is the wrong paper, we cannot see how the reference was validated, how calibration was checked, or whether the methods were tested fairly. Minor note: most of those open-source tools come from the same i2Nav-WHU line, so the evaluation is partly in-house; that is not disqualifying, but it would be stronger with independent baselines.\n\nNone of this is fatal to the dataset itself. A GitHub release with real data, per-segment error analysis, and independent checks could make this a solid resource. But as an arXiv submission, this is not coherent—the abstract describes a dataset, the body describes a different model. That is a manuscript integrity problem a desk editor cannot ignore.\n\nMy recommendation: desk reject or send back for a corrected resubmission. When the actual paper appears with indoor-reference validation, it deserves a proper peer review. I would not cite it yet.","headline":"A promising dataset abstract buried under the wrong manuscript body; the ground-truth claim needs independent validation before anyone should trust the centimeter-level figure.","tokens_in":17416,"tokens_out":3378,"would_cite":false,"duration_ms":35247,"reading_group":"no","serious_thinker":"no","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The i2Nav-Robot dataset provides about 17 kilometers of indoor-outdoor, multi-sensor navigation recordings with centimeter-level, fully covered ground truth, produced by post-processed integrated navigation with a high-grade IMU.","keywords":["dataset","multi-sensor fusion","indoor-outdoor navigation","ground truth","LiDAR","millimeter-wave radar","IMU","UGV"],"falsifier":"Compare the provided ground-truth poses against independent loop-closure constraints or surveyed control points placed inside the indoor parking areas; if the discrepancy grows beyond the claimed centimeter level when GNSS is absent, the 'fully covered centimeter-level ground truth' claim is refuted.","tokens_in":16551,"feed_emoji":"🚗","tokens_out":4914,"duration_ms":47926,"temperature":0.7,"pith_summary":"The paper presents i2Nav-Robot, a large-scale dataset for multi-sensor fusion navigation spanning outdoor streets and indoor parking lots, with about 17 kilometers of recorded driving across ten sequences. Its central claim is that it supplies high-rate, fully covered, centimeter-level ground truth for the entire route, obtained by post-processing integrated navigation with a high-grade IMU rather than by placing surveyed markers. The dataset also provides accurately time-synchronized data from a broad sensor suite, including solid-state LiDARs, 4D millimeter-wave radar, stereo cameras, IMU, GNSS receivers, and wheel odometers. The authors demonstrate the dataset's utility by running 15 open-sourced multi-sensor fusion navigation methods on it. A sympathetic reader would care because the lack of such indoor-outdoor datasets with reliable ground truth is a bottleneck for developing fusion navigation algorithms for UGVs.","feed_headline":"17 km dataset tracks indoor-outdoor robot navigation at cm accuracy","feed_subtitle":"Ten sequences pair LiDAR, radar, cameras, GNSS, and IMU with centimeter-level ground truth.","key_machinery":"The load-bearing mechanism is the post-processed integrated navigation solution built on a high-grade IMU, which produces the reference trajectory used as ground truth, combined with online hardware synchronization and offline calibration that align all sensor timestamps. The dataset's claim to fully covered centimeter-level ground truth rests on this integrated solution continuing to deliver centimeter accuracy in GNSS-denied indoor sections.","core_discovery":"The discovery is a dataset that combines indoor-outdoor coverage, a heterogeneous sensor suite with hardware-synchronized timestamps, and reference trajectories that claim centimeter-level accuracy everywhere, including inside parking garages where GNSS is unavailable. The reference trajectories are generated by post-processed integrated navigation—tightly fusing a high-grade IMU with GNSS and, presumably, other aiding data—rather than by external motion-capture or surveyed ground control. The paper claims that this design yields fully covered ground truth with no gaps in the indoor-outdoor transition, and that evaluation of 15 open-source fusion algorithms shows the data quality supports na","pith_inferences":["If the centimeter-level claim holds indoors, the dataset could serve as a testbed for evaluating IMU error models and dead-reckoning algorithms under GNSS-denied conditions, since the reference is independent of the sensors being tested.","Adding surveyed control points or loop-closure constraints in indoor sections would independently validate the ground truth and strengthen the 'fully covered' claim.","The same vehicle and synchronization pipeline could be reused to collect repeated passes over the same route, enabling studies of trajectory repeatability and sensor drift over time.","The paper's reliance on a single high-grade IMU for indoor reference suggests that comparing against an independent reference (e.g., a second IMU or visual-inertial SLAM) would quantify the actual indoor accuracy."],"forward_implications":["Researchers can benchmark multi-sensor fusion navigation methods on a common indoor-outdoor dataset with a consistent reference frame.","The synchronized 4D radar and solid-state LiDAR data enable fusion research involving newer sensor types.","The ten sequences across diverse scenarios support study of GNSS signal degradation and re-acquisition in parking structures and streets.","The dataset can serve as a training and evaluation resource for learning-based navigation approaches that need dense ground truth.","The documented time-synchronization pipeline provides a template for other sensor-fusion data collection efforts."],"supporting_citations":[],"fun_headline_variants":["17 km indoor-outdoor robot dataset with cm-accurate ground truth","Multi-sensor fusion dataset aids indoor-outdoor robot navigation","Robot nav dataset: 10 sequences, 17 km, cm ground truth","Hardware-synced sensors in 17 km indoor-outdoor robot dataset"],"cache_read_input_tokens":2816,"weakest_assumption_plain":"The reference trajectory is assumed to hold centimeter-level accuracy in the GNSS-denied indoor sections, where the post-processed solution must rely primarily on the high-grade IMU without any independent checks such as surveyed control points or loop closures.","fun_headline_variants_meta":{"raw":{"variants":["17 km indoor-outdoor robot dataset with cm-accurate ground truth","Multi-sensor fusion dataset aids indoor-outdoor robot navigation","Robot nav dataset: 10 sequences, 17 km, cm ground truth","Hardware-synced sensors in 17 km indoor-outdoor robot dataset"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000886,"raw_usage":{"total_tokens":3683,"prompt_tokens":790,"completion_tokens":2893,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":534,"completion_tokens_details":{"reasoning_tokens":2815}},"tokens_in":534,"tokens_out":2893,"duration_ms":20787,"temperature":1.0,"reasoning_tokens":2815,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T19:52:50.629026+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compare the provided ground-truth poses against independent loop-closure constraints or surveyed control points placed inside the indoor parking areas; if the discrepancy grows beyond the claimed centimeter level when GNSS is absent, the 'fully covered centimeter-level ground truth' claim is refuted.","supporting_citations":[],"review_version":1}