{"id":"c5da7b4f-312c-4264-9b85-5d594a4de948","arxiv_id":"2412.03887","paper_version":4,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"MOANA provides seven maritime sequences with synchronized X-band and W-band radar, LiDAR, stereo images, GNSS, and radar/camera object labels, plus odometry benchmarks.","lead":"MOANA is a new maritime sensor dataset that combines X-band and W-band radar with LiDAR, stereo cameras, and GNSS, collected across seven boat trips in South Korea and Singapore. It is the first public dataset pairing both radar types at sea, giving navigation researchers a resource for testing odometry, mapping, and object detection in ports and open water.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Radar-to-radar extrinsic calibration in §3.2.1 is accepted without reported residual or independent validation; a yaw or range error would corrupt the cross-sensor fusion that motivates the dataset, so this is the most load-bearing uncertainty.","rationale":"The paper's central contribution is a claimed first-of-kind dataset: simultaneous X-band and W-band scanning radar in maritime environments, plus LiDAR, stereo, GNSS, and labels. I read that claim in good faith. The dataset has real independent support as a data contribution: Table 2 gives sensor models and resolutions, Section 4.1 documents the file structure, Figure 5 shows maps, and Section 5 provides baseline odometry. The strongest claim is not disproven by anything in the text. However, the usefulness of a multi-radar benchmark depends on the extrinsics that relate X-band, W-band, LiDAR, and camera data. Section 3.2.1 describes only a coarse pixel-overlap calibration with CAD verticals and never reports residual errors. This is the same load-bearing weakness the reader identified. I would not move to REJECT because a dataset can still be released and the raw data may be valid, but the conditionality is justified. The manuscript itself supports this conditionality in Section 4.1.4 (GNSS drift, refined ground truth deferred) and Section 5.1 (W-band odometry evaluated on only one sequence). My proposed check, control-point reprojection with reported residuals, would settle whether the transforms are trustworthy. If residuals are small, the concern is resolved; if not, the central fusion narrative is weakened. Therefore the reader's CONDITIONAL verdict is appropriate; I do not change it.","tokens_in":10720,"tokens_out":7208,"duration_ms":71038,"concrete_test":"Select one port and one island sequence with overlapping X-band and W-band coverage; place or identify three or more stationary corner reflectors or distinct terrain features with surveyed GNSS positions; apply the released Base2Xband and Base2Wband extrinsics to project these points into both radar images. Compute per-range-bin RMS reprojection residual in meters and in pixels of the coarser X-band image, and also compute the relative transform error by independently solving for the radar-to-radar transform from these control points. If the control-point residual exceeds one X-band pixel or if the independent transform differs from the released transform by more than the stated positional and angular uncertainty, the cross-sensor fusion claim is not yet supported. Report the same residual for the CAD-derived vertical components using vessel pitch and roll variations if possible.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central value of MOANA is that X-band and W-band radar are jointly usable for maritime odometry and mapping. That joint usability rests on the extrinsic transforms in Section 3.2.1, which are estimated by resampling W-band polar images to X-band Cartesian resolution and aligning 'prominent features at the pixel level,' with vertical components taken from a CAD model. No residual errors, per-sequence validation, or uncertainty estimates are reported for these transforms, and the calibration section does not state how many features or independent views were used. The risk is concrete: a small yaw or translation error in Base2Xband or Base2Wband produces a cross-range error that grows with distance; by W-band maximum range (about 600 m) even a 0.5 degree yaw error is roughly 5 m, while X-band pixels are about 2.44 to 3.25 m. If the true transforms differ from the released ones by this order, the overlap maps in Figure 5, the fused odometry rationale, and any future multi-radar benchmark would inherit the misalignment. The paper's own Section 4.1.4 acknowledges GNSS drift and defers a refined ground truth, so there is no independent check in the paper that would catch a calibration bias. This is a limitation in evidence quality, not a demonstrated error, but it is the premise on which the dataset's claimed complementarity depends.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces MOANA, a maritime navigation dataset combining X-band marine radar, W-band scanning radar, LiDAR, stereo cameras, and GNSS, collected across seven sequences in port and island environments in South Korea and Singapore. The dataset includes extrinsic calibration files, GNSS-based trajectories, and 2D bounding-box object labels for one sequence (Single Island). The authors also report radar odometry benchmarks using CFEAR for W-band and LodeStar for X-band, with absolute trajectory error (ATE) results in Table 3. The central contribution claimed is that MOANA is the first multi-radar (X-band plus W-band) maritime dataset, enabling complementary short- and long-range perception for berthing, sailing, and docking scenarios.","tokens_in":10913,"tokens_out":3130,"duration_ms":29618,"significance":"If the calibration and ground truth are trustworthy, MOANA addresses a genuine gap: existing maritime radar datasets such as Pohang Canal provide only X-band data, while W-band radar datasets are predominantly land-based. The combination of a long-range X-band radar and a high-resolution short-range W-band radar, along with LiDAR, cameras, and object labels, could support research in maritime odometry, SLAM, place recognition, and object detection. The paper also provides a public website, ROS-based data publisher, and a per-sensor file structure, which are practical strengths for community adoption. However, the dataset's value hinges on the accuracy of the multi-radar extrinsic calibration and on the reliability of the GNSS-based reference, both of which currently lack quantitative validation. The benchmark results are suggestive but rest on a single W-band sequence and on an X-band method from the same group, so the empirical support for the complementarity claim is limited.","major_comments":[{"comment":"The radar-to-radar extrinsic calibration is a load-bearing component of the dataset's central claim, but the manuscript reports no residuals, no number of feature correspondences, no per-sequence validation, and no uncertainty estimates for the estimated x, y, and yaw parameters. The vertical components are taken from a CAD model. Given that X-band pixels are 2.44 m to 3.25 m and W-band range extends to 600 m, even a small yaw error (e.g., 0.5 degrees) produces a cross-range offset of several meters at the W-band maximum range, which would corrupt the overlap maps in Figure 5 and any future multi-radar fusion. The paper should provide independent validation of the extrinsic transforms, for example by comparing radar-projected features against LiDAR or satellite imagery in overlapping regions, or by reporting translation and rotation residuals from the feature alignment procedure for each sequence.","section":"§3.2.1"},{"comment":"The GNSS ground truth is acknowledged to suffer from signal reception failures and significant positional drift, and the authors explicitly defer a refined ground truth to future work. Table 3 reports ATE values computed against this drifting reference, yet no uncertainty, standard deviation, or per-sequence trajectory quality measures are given. This means the benchmark conclusions—such as 'W-band radar demonstrates lower error in the Near Port sequence' and 'X-band radar exhibits robust performance across all other sequences'—are not quantitatively substantiated. The authors should either release a corrected ground truth, report uncertainty bounds for the ATE values, or clearly label the numeric results as preliminary and support them with qualitative trajectory comparisons.","section":"§4.1.4 and Table 3"},{"comment":"The odometry benchmark covers only one of seven sequences for W-band radar (Near Port), and the W-band results for the other sequences are omitted because of blank images or tracking loss. The paper's broader claim that 'a complementary sensor configuration enables robust navigation' is therefore not directly demonstrated by the benchmark; it is inferred from the failure modes of each sensor alone. A simple fusion experiment (e.g., switching between W-band and X-band odometry based on range or feature availability) would provide direct evidence for the complementarity claim, or the wording should be softened to indicate that complementarity is a design rationale rather than a demonstrated result.","section":"§5.1 and Table 3"}],"minor_comments":[{"comment":"There is a typo in the first sentence: 'GNSS Reciever' should be 'GNSS Receiver.'","section":"§4.1"},{"comment":"The sentence 'The base for the yacht is established using the GNSS data. all the extrinsic calibration data are included in the calibration.' has a capitalization error and reads awkwardly; please revise for clarity.","section":"§3.2"},{"comment":"The table lists ATE values without units; please specify that the values are in meters and add a column or footnote indicating the uncertainty (e.g., standard deviation) for each trajectory.","section":"Table 3"},{"comment":"The X-band odometry method LodeStar is authored by the same group that presents MOANA. To help readers interpret the benchmark, the paper should explicitly note this potential bias and, if possible, include an independent X-band odometry baseline.","section":"§5.2"},{"comment":"The text says 'The first sequence, Port, was captured using a small fishing boat, while the Island sequence was recorded aboard a larger yacht,' but the dataset contains multiple sequences under Port and Island; please clarify that the two hardware configurations correspond to the two regional setups, not to individual sequences.","section":"§3.1"},{"comment":"The subsection numbering for the Island sequences is inconsistent: 'Complex Island' is listed as '(iii)' and 'Island Expedition' is also labeled '(iii)'; the latter should be '(iv).'","section":"§4.2.2"},{"comment":"The manual refinement of X-band range estimates using satellite imagery, LiDAR, and W-band radar data is described only briefly; more detail on how this refinement was performed and validated would help users understand the accuracy of the range calibration.","section":"§4.1.1"},{"comment":"The dataset is currently hosted on a Google Sites page; for long-term accessibility and citability, a DOI or a permanent archive (e.g., Zenodo) would be preferable.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is a dataset paper whose central contribution is plausible and useful, but the current evidence does not fully verify the calibration accuracy that the multi-radar fusion claim depends on. The benchmark also uses an X-band odometry method from the same group without an independent baseline, and the W-band evaluation is limited to one sequence. I recommend major revision rather than rejection because the dataset itself appears valuable and the calibration and ground-truth issues are addressable with additional validation or clearly stated limitations. The editor may also wish to consider whether the journal's standards for dataset papers require per-sequence calibration residuals and uncertainty measures for benchmark metrics."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"MOANA is a useful dataset paper. The contribution is real: no earlier public maritime dataset pairs X-band and W-band radar with LiDAR, stereo camera, and GNSS. Seven sequences span structured port and unstructured island environments, and one sequence carries 2D bounding boxes for radar and camera. The sequence variety and the GNSS-based maps give a good sense of what the different sensors do. The odometry baseline is a reasonable first step, even if thin.\n\nThe soft spots are all about evidence quality rather than the central claim. The radar-to-radar calibration in Section 3.2.1 is described in a paragraph: convert W-band polar to Cartesian, align prominent features at the pixel level, take vertical from CAD. No residuals, no validation, no number of views. The GNSS ground truth is acknowledged to drift, and refined ground truth is deferred. Table 3 lists ATE numbers with no uncertainty, and the W-band benchmark runs on a single sequence. These make the numbers hard to interpret but do not contradict the dataset's value.\n\nI'd also like to see an archival link with a version hash instead of a website; currently independent verification depends on the authors' site staying available.\n\nOverall, this deserves a serious referee. The dataset fills a concrete gap, the sensor configuration is sensible, and the limitations are stated honestly. A reviewer should push for calibration validation (even a sanity check on a known target), ground truth handling, and more comprehensive benchmarks before publication. The self-citation with LodeStar is fine; it is the authors' own method and the comparison is appropriate.","headline":"A genuinely useful first maritime X-band/W-band radar dataset; the main weakness is thin calibration evidence, not the central claim.","tokens_in":11543,"tokens_out":2032,"would_cite":true,"duration_ms":18726,"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":"The MOANA dataset claims to be the first maritime multi-radar dataset, pairing long-range X-band and short-range W-band radar to cover berthing, sailing, and docking.","keywords":["Dataset","Maritime","Radar","LiDAR","Object Detection","Place Recognition","Odometry","SLAM"],"falsifier":"Take any berthing segment where both radars and LiDAR see the same static structures, project the W-band radar and LiDAR into X-band coordinates using the provided extrinsics, and measure the pixel-level alignment of those structures; a systematic offset larger than a few pixels or degrees would break the complementary-fusion claim.","tokens_in":10471,"feed_emoji":"🌊","tokens_out":5115,"duration_ms":49295,"temperature":0.7,"pith_summary":"This paper introduces MOANA, a public maritime navigation dataset built around two radar bands with complementary strengths: X-band marine radar for long-range detection over kilometers and W-band imaging radar for high-resolution detection within about 600 meters. The dataset also includes LiDAR, stereo camera images, GNSS poses, and 2D bounding-box labels for one island sequence, across seven sequences in port, island, and seaside environments. The authors' central claim is that this multi-radar configuration is the first of its kind in maritime settings and that fusing the two bands gives more robust navigation than either sensor alone. If the claim holds, researchers gain a shared benchmark for radar odometry, SLAM, place recognition, and object detection in conditions where cameras and LiDAR are unreliable.","feed_headline":"First maritime dataset combines X-band and W-band radar for navigation","feed_subtitle":"Seven sequences from ports and islands let researchers benchmark radar odometry, SLAM, and object detection at sea.","key_machinery":"The load-bearing object is the dataset itself, with its synchronized multi-sensor recordings and calibration files. X-band data are Cartesian PNG images with ranges up to 2,498 m (Port) or 3,328 m (Island); W-band data are polar images converted to 1,024 by 1,024 Cartesian images at about 0.175 m per pixel over 600 m; LiDAR comes as point clouds; GNSS poses define the base frame; and one sequence carries 2D bounding-box labels for vessels and buoys. Calibration between the two radars is done by converting W-band polar images to Cartesian and matching prominent pixel-level features, with vertical extrinsics taken from a CAD model; LiDAR-to-W-band alignment uses phase correlation between polar images.","core_discovery":"On the paper's own terms, the discovery is that X-band and W-band radar provide complementary maritime perception: the W-band radar achieves lower odometry error in the near-harbor Near Port sequence (20.35 m absolute trajectory error with the CFEAR method, versus 63.48 m for X-band with LodeStar), while the X-band radar remains usable across all sequences where the W-band radar sees mostly empty water. The dataset is the first to record both radar bands together in a maritime environment, and the benchmark results are offered as evidence that a hybrid approach—short-range high-resolution W-band plus long-range X-band—can support the full range of vessel navigation, from berthing to open-water transits.","pith_inferences":["The paper does not report calibration residuals or uncertainties, so a user fusing all modalities should independently validate the provided extrinsics before trusting cross-sensor maps.","Because the two vessel setups use different W-band radar models (one with horizontal rays, one with downward-tilted rays), algorithms trained or tuned on Port sequences may need adjustment before they transfer to Island sequences.","The stated future work of refined GNSS/radar ground truth suggests that odometry evaluations on the current release will carry GNSS drift in regions of signal loss; early adopters may want to build their own reference trajectories for those stretches.","One testable extension is to run existing W-band place-recognition methods on the Island sequences to see whether natural features such as trees and rocks give repeatable radar descriptors, which the paper leaves open."],"forward_implications":["Radar odometry and SLAM methods can be benchmarked in maritime settings with both long- and short-range sensors on the same platform.","W-band radar can be treated as a LiDAR alternative for near-field maritime perception where fog, salt spray, and corrosion degrade optical sensors.","Fusing X-band and W-band radar should maintain navigation continuity in open water and during berthing, where each band alone fails.","The labeled Single Island sequence provides a testbed for radar-based object detection under multipath ghosting and occlusion.","Sequence overlaps support inter- and intra-sequence place recognition experiments."],"supporting_citations":[{"why":"The prior maritime radar dataset with X-band only; MOANA positions itself as its multi-radar extension.","marker":"Chung et al. (2023)"},{"why":"Established W-band radar as a navigation sensor on ground vehicles, the capability MOANA moves to maritime settings.","marker":"Barnes et al. (2020)"},{"why":"Provided W-band radar place-recognition methods and the phase-correlation calibration approach reused for LiDAR-to-W-band alignment.","marker":"Kim et al. (2020)"},{"why":"Survey that motivates W-band millimeter-wave radar as a robust alternative to LiDAR, supporting the sensor choice.","marker":"Harlow et al. (2024)"},{"why":"The CFEAR W-band radar odometry method used for the near-port benchmark.","marker":"Adolfsson et al. (2022)"},{"why":"The LodeStar X-band radar odometry method used for benchmarks across all sequences.","marker":"Jang et al. (2024)"}],"fun_headline_variants":["First maritime dataset pairs X and W-band radar for navigation","X-band and W-band radar join in new maritime odometry dataset","Dataset benchmarks radar odometry at sea with dual-band coverage","Maritime radar dataset combines near and far field for SLAM","New maritime dataset fuses X and W-band radar for robust nav"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The dataset's claimed value for sensor fusion rests on the accuracy of the extrinsic calibration among X-band radar, W-band radar, LiDAR, and cameras, yet the paper reports no residual errors and takes vertical alignments from a CAD model.","fun_headline_variants_meta":{"raw":{"variants":["First maritime dataset pairs X and W-band radar for navigation","X-band and W-band radar join in new maritime odometry dataset","Dataset benchmarks radar odometry at sea with dual-band coverage","Maritime radar dataset combines near and far field for SLAM","New maritime dataset fuses X and W-band radar for robust nav"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000692,"raw_usage":{"total_tokens":3142,"prompt_tokens":967,"completion_tokens":2175,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":583,"completion_tokens_details":{"reasoning_tokens":2089}},"tokens_in":583,"tokens_out":2175,"duration_ms":14282,"temperature":1.0,"reasoning_tokens":2089,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T21:58:07.790124+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take any berthing segment where both radars and LiDAR see the same static structures, project the W-band radar and LiDAR into X-band coordinates using the provided extrinsics, and measure the pixel-level alignment of those structures; a systematic offset larger than a few pixels or degrees would break the complementary-fusion claim.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The prior maritime radar dataset with X-band only; MOANA positions itself as its multi-radar extension."},{"cited_title":"In: Proc","cited_arxiv_id":null,"evidence_quote":"Established W-band radar as a navigation sensor on ground vehicles, the capability MOANA moves to maritime settings."},{"cited_title":"In: Proc","cited_arxiv_id":null,"evidence_quote":"Provided W-band radar place-recognition methods and the phase-correlation calibration approach reused for LiDAR-to-W-band alignment."},{"cited_title":"IEEE Transactions on Robotics","cited_arxiv_id":null,"evidence_quote":"Survey that motivates W-band millimeter-wave radar as a robust alternative to LiDAR, supporting the sensor choice."},{"cited_title":"IEEE Transactions on robotics 39(2): 1476--1495","cited_arxiv_id":null,"evidence_quote":"The CFEAR W-band radar odometry method used for the near-port benchmark."}],"review_version":1}