{"id":"e0a6b6b3-15b0-4689-bded-6c9a892b4eb9","arxiv_id":"1908.02827","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Three riverine coverage algorithms for autonomous surface vehicles, including an equal-triangle zigzag heuristic, are described and tested in simulation and on field trajectories.","lead":"This paper presents three deterministic coverage planners for autonomous boats on rivers, one longitudinal, one zigzag, and one transversal, and validates them in simulation and on the Congaree River. A generalist might read it to see how standard robotic coverage ideas are adapted to a practical environmental monitoring task with real field data.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The completeness claim for L-Cover/T-Cover is asserted without proof and is undercut by the paper's own Table I, which reports only 92.65%/91.42% covered area, not 100%.","rationale":"The reader's weakest assumption about a constant bathymetric sensor footprint is a real limitation and is explicitly acknowledged in the paper. However, it is not the deepest problem with the central claim. Even granting a fixed footprint, the paper does not prove that the generated L-Cover and T-Cover paths leave no gap, and the reported simulation coverage percentages (92.65% and 91.42%) are inconsistent with a complete coverage guarantee under the paper's own swath-proportional-to-spacing metric. A curved river exposes a concrete mechanism for failure: transverse passes spaced along one shore or centerline become more than s apart at the other shore, so a fixed-width swath cannot cover the outer-bank wedge. This is an internal correctness risk rather than a disagreement with external consensus. The paper still has value as an empirical comparison of realistic riverine survey patterns, and the completeness concern is addressable by adding a formal spacing condition or by re-running experiments with swath width equal to spacing. Therefore the appropriate verdict remains CONDITIONAL rather than ACCEPT or REJECT, but the central claim should not be taken as established without further evidence.","tokens_in":8208,"tokens_out":7135,"duration_ms":86155,"concrete_test":"Implement T-Cover on a constant-width circular-arc river with inner radius R, width W, and set the pass spacing s equal to the sensor swath width. Rasterize the region of interest at fine resolution and compute the set of cells not within distance s/2 of any point on the planned path. If any uncovered cells appear, especially near the outer bank, then the universal completeness claim fails under the paper's own fixed-footprint assumption.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The paper's central claim is that L-Cover and T-Cover guarantee complete coverage: 'ensuring that for a fixed sensor footprint no area remains uncovered' (Introduction). This is stated without a proof, and the paper's own quantitative evidence is in tension with it. Section IV-A defines coverage by assuming 'the travel path pi has a width proportional to the spacing value s.' Under that definition, if the proportionality constant equals 1, a gap-free planner should cover essentially 100% of the region of interest. Table I reports Area Covered = 92.65% for L-Cover and 91.42% for T-Cover, not 100%. Thus either the simulations never instantiated the condition under which the completeness guarantee is claimed, or the guarantee is false for the generated paths. The geometry of T-Cover also makes the claim non-obvious: transverse passes 'spaced by distance s from each other' (Section III-C) cannot be separated by exactly s along both banks simultaneously when the river is curved. If spacing is fixed along one bank or along the centerline, the separation between adjacent transverse passes exceeds s near the outer bank, leaving wedge-shaped uncovered regions. The same risk arises in L-Cover when clusters with varying width are merged (Algorithm 1, Lines 13-15). No geometric argument in the paper rules out these gaps. The strongest claim therefore rests on an unstated condition relating path spacing to sensor footprint in curved, width-varying rivers, and that condition is not established by the reported data.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper addresses autonomous coverage path planning for riverine surveying with an autonomous surface vehicle (ASV). Three deterministic planners are proposed: L-Cover, which runs longitudinal passes parallel to the shores and adapts the number of passes to the river width; Z-Cover, a zigzag partial-coverage strategy intended for single-pass surveys; and T-Cover, a transverse lawn-mowing strategy. The authors claim that L-Cover and T-Cover provide complete coverage for a fixed sensor footprint, and they report simulation results on real river maps as well as field deployments on the Congaree River, stating that the approach improves accuracy and efficiency over manual surveying.","tokens_in":8541,"tokens_out":2967,"duration_ms":34977,"significance":"If the central claims are supported, the work is a useful practical contribution to an application domain where autonomous riverine surveying is still uncommon. The paper gives explicit algorithmic descriptions, compares against a fixed-angle manual surveying baseline, and includes real field deployments with bathymetric and side-scan sonar data, which is commendable. The main significance, however, rests on the completeness guarantee for L-Cover and T-Cover, and on the quantitative superiority over human performance. These claims are not adequately supported by the paper's own data: the reported coverage percentages are well below 100%, the reported field distances do not match the abstract's 35 km figure, and the 'accuracy' improvement lacks quantitative evidence. The practical value of the methods is plausible, but the paper as written does not substantiate its headline claims.","major_comments":[{"comment":"The completeness claim is in direct tension with the paper's own simulation results. The Introduction states that L-Cover and T-Cover ensure that 'for a fixed sensor footprint no area remains uncovered,' but Table I reports Area Covered of 92.65% for L-Cover and 91.42% for T-Cover. Since §IV-A defines coverage by giving the travel path a width proportional to the spacing parameter s, a gap-free planner should cover close to 100% of the region of interest under the assumed swath model. The authors should either provide a formal proof or a precise set of conditions under which the completeness guarantee holds, or reconcile the metric with the reported percentages. Without this, the paper's strongest theoretical claim is unsupported.","section":"§I and §IV-A, Table I"},{"comment":"The abstract and introduction state that the field deployments produced 'more than 35km of coverage trajectories,' but Table II lists Total Distance values of 5.2 km (Z-Cover), 10 km (T-Cover), and 13.02 km (L-Cover), which sum to 28.22 km. Even the Coverage Distance column sums to only 21.32 km. If additional deployments were performed beyond those listed in Table II, they must be reported; if not, the 35 km figure appears to be an arithmetic inconsistency. This discrepancy undermines confidence in the quantitative reporting and should be fixed.","section":"Abstract and §IV-B, Table II"},{"comment":"The fixed-footprint assumption is load-bearing for the completeness claim but is neither derived nor tested. The paper states that 'the footprint of the bathymetric sensor ... is constant and can be calculated based on the average depth of the area/river.' However, L-Cover and T-Cover generate paths in curved, width-varying rivers: in T-Cover, transverse passes spaced by distance s along one bank or centerline can be separated by more than s near the outer bank of a bend, leaving wedge-shaped uncovered regions, and in L-Cover the cluster-merging step in Algorithm 1 (Lines 13-15) can produce similar gaps. The paper provides no geometric argument ruling out such gaps. At minimum, the authors should state the exact conditions under which the guarantee holds and quantify gap size empirically for the simulated and field environments.","section":"§IV-B and §III-C"},{"comment":"The claimed 'increases in accuracy and efficiency compared to human performance' are not quantitatively supported. Table II provides times and distances, but the accuracy comparison is qualitative; Figure 5 shows depth and uncertainty maps, but no numerical RMSE, coverage error, or statistical comparison against manual surveys is reported. The paper should either add quantitative accuracy metrics (e.g., RMSE values from the GP maps, or comparison of bathymetric estimates) or explicitly limit the claim to efficiency and qualitative map quality.","section":"§IV-B, Figure 5, and Abstract"}],"minor_comments":[{"comment":"The pseudocode for Z-Cover is ambiguous: the unconditional 'break' after the triangle-area check appears to exit the for loop after the first candidate, making the retry logic in Lines 13-15 unreachable. Please revise the indentation and control flow to match the intended search over d candidate lines.","section":"Algorithm 2"},{"comment":"There are several typographical errors, including 'a a lawn-mowing pattern' in Section III-C, 'Figure Figure 2' in Section III-A, and 'parameters describing' in Section III-A. A careful proofreading pass is recommended.","section":"Throughout"},{"comment":"The table lists algorithms in the order Z-Cover, T-Cover, L-Cover, while the text in Section IV-B discusses them in a different order. Please align the order for readability.","section":"Table II"},{"comment":"The Z-Cover method is described as producing triangles with approximately equal areas, but the relation between equal triangle areas and 'the ratio of the covered areas across the river' is not explained. A short derivation or diagram reference would help the reader understand the geometric motivation.","section":"§III-B"}],"recommendation":"major_revision","confidential_remarks":"The manuscript addresses a worthy application area and the field experiments are valuable, but the central completeness claim and the quantitative accuracy claims need substantial revision. The 35 km versus 28.22 km discrepancy in particular should be caught by the authors before resubmission; it makes the reader question the reliability of other reported numbers. I would encourage the authors to provide an explicit geometric analysis of the spacing condition for curved rivers and to include numerical accuracy metrics, as these are the load-bearing points of the paper."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The most useful piece is the Z-Cover equal-triangle heuristic and, to a lesser degree, L-Cover's width-adaptive clustering. The field deployments on the Congaree with a real ASV are a genuine plus; the qualitative backscatter comparison showing the autonomous L-Cover covering roughly twice the area of a manual run in the same time is worth seeing.\n\nThat said, the central promise is not delivered. The introduction says L-Cover and T-Cover ensure no area remains uncovered for a fixed sensor footprint, but Table I reports 92.65% and 91.42% covered area, and the coverage metric is computed by giving the generated path a width proportional to the planner's own spacing parameter. That makes the percentage a property of the path, not an independent measurement. No geometric argument rules out gaps in curved rivers: T-Cover passes spaced by s along the centerline or one bank will separate by more than s near the outer bank, leaving wedges. The stress-test note is on target here.\n\nThere are also presentation problems a referee would catch: the abstract promises more than 35 km of coverage trajectories, but Table II sums to 28.22 km total traveled (21.32 km of coverage distance). Algorithm 2's pseudocode has an unconditional break inside the for loop, so as printed it only ever tests the first candidate line; that looks like a bug. Table I has no error bars, so the 1.7% difference between Z-Cover variants is not clearly meaningful. The constant-footprint assumption is stated explicitly, but it is load-bearing; if the side-scan footprint varies with depth, the completeness claim and the percentages both change.\n\nThe core engineering is plausible, and the paper is honest about the manual-practice inspiration. The flaws are addressable but they matter: either revise the completeness claim to 'attempts complete coverage' and explain the coverage gap, or provide a real proof and tune the simulator. Fix the pseudocode, reconcile the field totals, and report variance across the simulated trials.\n\nI'd send it to peer review with a request for major revision. The Z-Cover idea and the field data deserve referee time, but the paper in its current form overstates what it shows.","headline":"A real field robot and a new zigzag heuristic, but the paper's completeness guarantee is undercut by its own Table I and the pseudocode has a bug.","tokens_in":9069,"tokens_out":4739,"would_cite":false,"duration_ms":45745,"reading_group":"maybe","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Autonomous river-survey boats can achieve complete coverage with shore-parallel or shore-perpendicular passes, and a bank-to-bank zigzag beats the fixed-angle method now used by human surveyors.","keywords":["coverage path planning","autonomous surface vehicle","riverine surveying","bathymetric mapping","complete coverage","zigzag coverage","side-scan sonar","field robotics"],"falsifier":"Run L-Cover or T-Cover over a reach with a known deep channel and shallow bar, georeference the actual sonar returns, and compare the measured uncovered riverbed with the prediction made using a footprint width from average depth; any gap wider than the assumed swath refutes the completeness claim.","tokens_in":8037,"feed_emoji":"🚤","tokens_out":7290,"duration_ms":76643,"temperature":0.7,"pith_summary":"This paper addresses the practical problem that rivers are still surveyed mostly by manually piloted boats, and that standard lawn-mower coverage planners waste motion in narrow meandering waterways. It proposes three deterministic planners that encode the strategies human surveyors actually use: L-Cover, long shore-parallel passes; T-Cover, shore-perpendicular lawn-mowing passes; and Z-Cover, a single bank-to-bank zigzag. The central claim is that L-Cover and T-Cover give complete coverage for a fixed sensor footprint, while Z-Cover samples the full width in one pass and distributes samples more evenly than the fixed-angle zigzag baseline used in practice. Simulation on real river maps and field deployments totaling more than 35 km on the Congaree River support the claim, with L-Cover covering about twice the area of a manual run in comparable time and producing cleaner bathymetric mosaics.","feed_headline":"Two boat paths guarantee complete river coverage","feed_subtitle":"Field runs covered 35 km and produced cleaner bathymetry than manual boating.","key_machinery":"The load-bearing mechanism is the directional contour representation of the river, an ordered list of shore points that lets the planner reason about width and downriver direction. That representation drives three geometric routines: L-Cover's width-based clustering and parallel-pass generation, Z-Cover's equal-triangle area selection, and T-Cover's shore-perpendicular decomposition. The machinery converts coverage into a spacing problem: with a constant pass spacing and a fixed sensor footprint, covering every width-homogeneous cluster with the correct number of passes leaves no area uncovered.","core_discovery":"On the paper's own terms, riverine coverage is a geometric partitioning problem. L-Cover divides the river into clusters whose widths are close enough that a constant number of shore-parallel passes, spaced by a parameter s, covers each cluster completely; the number of passes adapts to the width. T-Cover instead lays passes perpendicular to the shores, applying the boustrophedon idea along the river's curvature. Z-Cover chooses each next shore-contact point so that consecutive triangles formed with the previous two path points have nearly equal areas, which spreads samples evenly across the river rather than overshooting one bank. The reported simulation numbers are 92.65% covered area for L-Cover, 91.42% for T-Cover, 31.05% for Z-Cover, and 29.39% for the fixed-angle heuristic; in the field, an L-Cover trajectory covered roughly twice the area of a manual survey in about the same operating time.","pith_inferences":["A natural next step, not taken in the paper, is to replace the constant-footprint assumption with a depth-dependent sonar model and let pass spacing adapt locally; the completeness guarantee would then extend to rivers with strong bathymetric relief.","The equal-triangle rule in Z-Cover could be made flow-aware by weighting triangle areas with expected current drift, which would likely reduce the overshoot observed in fast-moving reaches.","The same width-clustering idea transfers to other corridor environments, such as narrow aerial canyons or underwater channels, where ordinary lawn-mower decompositions pay a high penalty in turns."],"forward_implications":["L-Cover is the preferred complete-coverage pattern for side-scan sonar surveys because it couples high covered area (92.65% in simulation) with a short return path (8.9%).","T-Cover reaches similar completeness (91.42%) but with more turns and a longer return trip, so it fits missions where the riverbed must be sampled across the width within a short time window.","Z-Cover is a partial-coverage method: it samples about a third of the river in one pass, with slightly better and more even coverage than the fixed-angle heuristic while avoiding severe overshoot.","The choice of planner affects map quality, not just path length: bathymetric maps built from L-Cover data showed lower uncertainty than those from Z- or T-Cover in the field trials.","More than 35 km of executed trajectories show the planned paths are trackable by GPS waypoint navigation on real water, so the geometric guarantees survive practical deployment."],"supporting_citations":[{"why":"Supplies the boustrophedon cellular decomposition that T-Cover's transverse lawn-mowing pattern is based on.","marker":"[1]"},{"why":"Describes the autonomous surface vehicle used for the field deployments and coverage trials.","marker":"[4]"},{"why":"Provides the fixed-angle zigzag surveying baseline against which Z-Cover's equal-triangle method is compared.","marker":"[8]"},{"why":"Supplies the simulation environment used to test all three planners on maps of real rivers.","marker":"[19]"},{"why":"Provides the Gaussian-process mapping technique used to turn the collected sonar readings into bathymetric maps.","marker":"[22]"}],"fun_headline_variants":["Autonomous boat doubles river area surveyed per run","Field-tested boat paths cover twice the river area","35 km covered by autonomous river survey bots","Autonomous boat coverage beats human accuracy and speed","Robot boat sweeps rivers with complete coverage paths"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The completeness guarantee depends on the sensor footprint being a fixed width computed from the average river depth; if depth, sensor tilt, or turning motion changes the footprint, the planner can leave gaps that its coverage percentages do not predict.","fun_headline_variants_meta":{"raw":{"variants":["Autonomous boat doubles river area surveyed per run","Field-tested boat paths cover twice the river area","35 km covered by autonomous river survey bots","Autonomous boat coverage beats human accuracy and speed","Robot boat sweeps rivers with complete coverage paths"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000846,"raw_usage":{"total_tokens":3646,"prompt_tokens":875,"completion_tokens":2771,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":491,"completion_tokens_details":{"reasoning_tokens":2700}},"tokens_in":491,"tokens_out":2771,"duration_ms":24216,"temperature":1.0,"reasoning_tokens":2700,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:31:53.470478+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run L-Cover or T-Cover over a reach with a known deep channel and shallow bar, georeference the actual sonar returns, and compare the measured uncovered riverbed with the prediction made using a footprint width from average depth; any gap wider than the assumed swath refutes the completeness claim.","supporting_citations":[{"cited_title":"Morse Decompositions for Coverage Tasks,","cited_arxiv_id":null,"evidence_quote":"Supplies the boustrophedon cellular decomposition that T-Cover's transverse lawn-mowing pattern is based on."},{"cited_title":"An autonomous surface vehicle for long term operations,","cited_arxiv_id":null,"evidence_quote":"Describes the autonomous surface vehicle used for the field deployments and coverage trials."},{"cited_title":"Two-dimensional mapping and tracking of a coastal upwelling front by an autonomous underwater vehicle,","cited_arxiv_id":null,"evidence_quote":"Provides the fixed-angle zigzag surveying baseline against which Z-Cover's equal-triangle method is compared."},{"cited_title":"Massively multi-robot simulation in stage,","cited_arxiv_id":null,"evidence_quote":"Supplies the simulation environment used to test all three planners on maps of real rivers."},{"cited_title":"Rasmussen and C","cited_arxiv_id":null,"evidence_quote":"Provides the Gaussian-process mapping technique used to turn the collected sonar readings into bathymetric maps."}],"review_version":1}