REVIEW 4 major objections 6 minor 1 references
Near-Range Environmental Perception for Inland Waterway Vessels: A Comparative Study of LiDAR and Automotive FMCW RADAR Sensors
T0 review · 4 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash
Pith's one-line read A distributed network of automotive FMCW radars can perceive close-range static features around an inland waterway vessel, the paper argues, despite producing far sparser point clouds than LiDAR.
desk verdict Useful field dataset with a credible weak claim, but the environment comparison is confounded by different sensor setups and the density metric needs fixing. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The central object is the automotive FMCW radar sensor network itself: four compact 76 to 81 GHz radars mounted around the boat, each using Doppler division multiple access (DDMA) with three transmit and four receive antennas to form 12 virtual receive antennas. The processing chain converts the raw data cube via fast-time and slow-time Fourier transforms into range-Doppler maps, then a further antenna-dimension Fourier transform yields angle of arrival, producing a 3D point cloud with range, angle, signal strength, and radial velocity. The comparative evaluation rests on two quantitative tools: the mean cell count per square meter, measured on a 100 by 80 meter grid, and the Jaccard coefficient, which measures spatial agreement between LiDAR and radar occupancy cells.
What would settle it
Conduct the same two-environment comparison with an identical sensor configuration, meaning the same number and placement of radar units, the same LiDAR model, the same reference frame, and the same time synchronization, while controlling for weather. If, under that controlled replication, the radar point clouds no longer agree with LiDAR on static features such as bridge undercrossings and lock walls, or if the radar's scenario-independence disappears and becomes purely a hardware difference, then the paper's promise for radar-based inland waterway perception would be refuted.
Extended reading notes
Core claim
The discovery this paper reports is that a network of four automotive FMCW radars, operating at 76 to 81 GHz, can perceive close-range static environmental features around an inland waterway vessel, and that its detection behavior is less dependent on the environment than LiDAR's. In the city scenario the average LiDAR point density per square meter was 30 to 50 times higher than the radar's, but in the natural scenario the LiDAR advantage shrank to roughly a factor of three because vegetation blocked the narrow light beams. The Jaccard similarity between LiDAR and radar occupancy grids rose sharply when highly reflective radar targets such as a ship or bridge guard rails appeared, indicating that the radar captures these static landmarks even when its point cloud is sparse.
Load-bearing premise
The comparison assumes that the two field campaigns can be treated as directly comparable, even though the Berlin setup used two frontal radars with LiDAR 1 while the Neustrelitz setup used four radars with LiDAR 2.
Editorial extensions
If this is right
- The tested radar network can serve as an all-weather complement to LiDAR for close-range maneuvers, since it keeps detecting static features where LiDAR is degraded by vegetation or poor visibility.
- Radar point clouds, though sparse, contain enough structural information to support radar-only SLAM and local positioning in inland waterway environments.
- The large density gap in favor of LiDAR in the city scenario indicates that radar data will need denser fusion or longer aggregation times when used alone in concrete-heavy infrastructure.
- The Jaccard similarity peaks on strongly reflective objects suggest that radar-specific landmarks such as guard rails, ship hulls, and bridge elements should be exploited as features for mapping and localization.
- Future gains in time synchronization, sensor placement measurement, and localization algorithms are prerequisites for turning the promising assessment into a deployable system.
Reading between the lines
- Because the two campaigns confound environment with hardware, the sharp city-to-nature difference in LiDAR's density advantage should be read as a hypothesis about vegetation effects, not a proven LiDAR weakness.
- If radar detection truly is more scenario-independent than LiDAR's, the same sensor network could simplify the sensor suite on inland vessels by replacing multiple LiDAR units with a smaller radar network plus a single camera for semantic context.
- A direct testable extension would be to run the same radar network and LiDAR simultaneously in rain or fog, predicting that radar Jaccard similarity to a reference map stays stable while LiDAR's falls.
- The density metric's sensitivity to cell size suggests that the practical comparison should be tied to the actual feature scale needed for docking and lock entry, not a fixed one-meter grid.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper presents an experimental evaluation of a network of automotive FMCW radar sensors for near-range environmental perception on an inland waterway vessel, with LiDAR as a reference. The authors describe the radar signal processing pipeline, the Aurora test platform, and two measurement campaigns (Berlin city and Neustrelitz nature). They compare radar and LiDAR point clouds using a density metric (Eq. 1) and the Jaccard coefficient (Eq. 2), and conclude that, despite lower point cloud density, the radar network is promising for perceiving close-range static features and appears more environment-independent than LiDAR.
Significance. If the claims were fully supported, this work would be a useful step toward all-weather close-range perception for automated inland navigation, where traditional marine radar has a blind zone and LiDAR degrades in poor visibility. The paper provides a real field dataset and a concrete processing chain, which are valuable assets for the community. However, the current quantitative comparison is weakened by a confounded experimental design and an inconsistently defined density metric, so the paper's broader comparative conclusions outrun the evidence.
major comments (4)
- [Sections 2.3 and 3] The cross-environment comparison is confounded: Berlin used two frontal radars and LiDAR 1, whereas Neustrelitz used four radars and LiDAR 2. The differences in the radar/LiDAR density ratio (roughly 3 versus 30–50) and in the LiDAR distribution are attributed to vegetation and the environment, but the number of radar sensors, their mounting positions, and the LiDAR model/range/resolution all changed simultaneously. In particular, a factor of two in radar target count is expected simply from using twice as many sensors. The statement in Section 3 that 'radar sensors appear to be more independent of the scenarios selected here' is not identifiable from these data and should be either removed or explicitly labeled as a hypothesis requiring a controlled experiment.
- [Section 2.4, Eq. (1)] The density metric is internally inconsistent and ambiguous. The text says the mean is taken over cells with N_m > 1, but the denominator M-K subtracts cells where N_m equals zero 'either for LiDAR or RADAR.' If K excludes cells that are empty for either sensor, the average is computed over a biased subset that favors cells where both sensors detect something, making the metric depend on the sensors' coverage overlap. Additionally, it is not clear whether N_m counts points from a single sensor or from the combined point cloud; the equation needs a precise definition of N_m, M, and K, and separate per-sensor densities should be reported with confidence intervals.
- [Section 3, Fig. 4] The lower LiDAR density in Neustrelitz is attributed to vegetation preventing light beams from passing through leaves, but a simpler hardware explanation is available: LiDAR 2 has a maximum range of 75 m versus 200 m for LiDAR 1, and a different horizontal/vertical resolution. A shorter-range sensor would naturally produce fewer and more centrally distributed detections in a 100 × 80 m crop. The authors do not address this alternative, so the 'centralized distribution' is not sufficient evidence for a vegetation effect.
- [Section 3, Fig. 5] The Jaccard analysis is described too loosely to support the statements made. The caption mentions 'parameterized cell size' and the text claims similarity 'decreases with reduced cell size,' but no cell sizes are listed in the text or caption, and the curves in Fig. 5 are not identified. Similarly, the peaks of similarity are linked to events such as a ship or guard rails, but the event annotations are not described or quantified. Without a clear protocol for cell-size variation and event labeling, the similarity discussion is largely qualitative.
minor comments (6)
- [Throughout] The manuscript contains numerous typographical errors and duplicated words, including 'therefore therefore' (Section 1), 'autmotive' (Section 4), 'Jaccarr' (Section 3), 'Neutrelitz' (Fig. 4), 'enviroment' (Section 1), 'maesurement' (Fig. 3 caption), 'fotograph' (Fig. 1 caption), 'devellope' (Section 2.2), 'simultanously' (Section 2.2), 'corrensponding' (Section 3), 'parameparameterized' (Fig. 5 caption), and 'secenarios' (Section 3). A thorough proofread is needed.
- [Section 3] The phrase 'Take into account Fig. ,' contains an incomplete figure reference; the intended figure or table number is missing.
- [Table 1] The LiDAR and RADAR specification table is difficult to read: the two velocity rows are not labeled separately, the range resolution values are split across lines and contain formatting artifacts such as '0 .08' and '1.561, 3', and the azimuth/elevation resolution entries are paired with footnote references that are only partially explained. The table should be reformatted with clear row labels and explicit footnotes.
- [Section 2.4, Eq. (1)] The summation index runs from m=0 to M, but if M is the total number of cells, the index should run from 1 to M; if M is an upper index in a zero-based array, the total number of cells is M+1. This should be clarified.
- [Section 2.1] The text states that 'the camera captures semantic data,' but semantic camera data is not used anywhere in the comparison; either remove this sentence or explain its relevance to the present study.
- [Section 2.3] The phrase 'TWe used two measurement campaigns' contains a typo ('TWe' should be 'We'). In addition, the scenes are described as representative samples, but the duration, number of frames, and weather conditions of each campaign are not reported, which limits reproducibility.
Circularity Check
No circularity found: the study is an experimental sensor comparison with independent LiDAR reference and no self-referential derivation or fitted prediction.
full rationale
This paper is an experimental, measurement-based comparison, not a derivation chain. The central claim that the automotive RADAR network can perceive close-range static environmental features is supported directly by superimposed LiDAR and RADAR point clouds and by the Jaccard coefficient (Eq. 2), which measures set overlap between independent sensor outputs. The density metric in Eq. (1) is an averaging definition applied to measured point counts, not a parameter fitted to the conclusion it is used to support. LiDAR is treated as an external reference, and temporal alignment is provided by the PNT/GNSS unit rather than by any quantity derived from the comparison itself. The only self-citation, [1], concerns the map processing unit to be used in future work and is not load-bearing for the perception claim. The acknowledged non-identical sensor setups between Berlin and Neustrelitz (two vs. four RADAR sensors and different LiDAR models) are a genuine experimental-design limitation that weakens the cross-environment comparison, but this is a validity concern about confounding variables, not circular reasoning: the paper does not define or predict the environment-dependence result from the same data in a way that makes it true by construction. Therefore, no circular step is present and the appropriate score is 0.
Assumptions & free parameters
assumptions (4)
- domain assumption Standard FMCW signal processing: FFT along fast-time yields range, FFT along slow-time yields Doppler, and FFT over antennas yields angle (Section 2.2).
- domain assumption The DDMA scheme with three transmitters and four receivers produces 12 independent virtual antennas (Section 2.2).
- domain assumption LiDAR point clouds can serve as a reference for the physical environment (throughout the comparison).
- domain assumption The PNT unit provides a sufficiently accurate common time and position reference for fusing radar and LiDAR point clouds (Section 2.1).
Cite this review
Pith. "Pith review of Near-Range Environmental Perception for Inland Waterway Vessels: A Comparative Study of LiDAR and Automotive FMCW RADAR Sensors." pith.science (2026). https://pith.science/paper/GTNYVHPU
@misc{pith2026241115901,
author = {Pith},
title = {Pith review of: Near-Range Environmental Perception for Inland Waterway Vessels: A Comparative Study of LiDAR and Automotive FMCW RADAR Sensors},
year = {2026},
howpublished = {\url{https://pith.science/paper/GTNYVHPU}},
note = {Machine review of arXiv:2411.15901}
}
read the original abstract
Advancing towards high automation and autonomous operations is crucial for the future of inland waterway transport (IWT) systems. These systems necessitate robust and precise onboard sensory technologies that can perceive the environment under all weather conditions, including static features for local positioning techniques such as Simultaneous Localization and Mapping (SLAM). Traditional marine RADAR, mandatory on vessels and operating in the 9300-9500 MHz frequency band, can cover ranges from 15 to 1200 meters but are inadequate for detecting closer objects, making them unsuitable for automated docking maneuvers, lock entry, or bridge undercrossings. This necessitates the development of reliable close-range sensor technology that functions effectively in all weather conditions. In present research works on vessel automation, LiDAR sensors, operating in the nearinfrared range, are used predominantly to detect the immediate surroundings of vessels but suffer significant degradation in poor visibility. Conversely, automotive RADAR sensors, utilizing the 76-81 GHz frequency band, can detect objects from a few centimeters to up to 200 meters, even in adverse conditions. These sensors are commonly used in advanced autonomous road traffic systems and are evaluated in this study for their suitability in inland navigation and maneuvering. This paper discusses a distributed sensor network of four compact automotive frequencymodulated continuous-wave (FMCW) radars mounted on a cabin boat as a test platform. Initial field experiments demonstrate the RADAR network's ability to perceive closerange static environmental features around the boat in inland waters. The paper also provides a comparative analysis of the environmental detection capabilities of automotive RADAR and LiDAR sensors.
Reference graph
Works this paper leans on
-
[1]
High definition mapping for inlandwaterways: Techniques, challenges and prospects
[1] L. H ¨osch, Y . Wellknown, A. Llorente, X. An, J. P. Llerena, and D. Medinaand, “High definition mapping for inlandwaterways: Techniques, challenges and prospects”, 2023 IEEE 26th Interna- tional Conference on Intelligent Transportation Systems (ITSC), Bilbao, Spain, 2023 pp. 6034-6041 9
work page 2023
Reviewed August 12, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.