{"id":"20271d56-6113-46c9-8796-1621541db7c4","arxiv_id":"2608.11641","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"A risk-aware online routing planner for truck-assisted drone delivery adds wind uncertainty margins and return-feasibility checks, improving simulated mission success rates under replayed wind logs.","lead":"A drone-and-truck delivery system uses live wind estimates with built-in safety margins to choose routes that keep enough battery for the drone to come back. In simulations with recorded wind data, the new planner completes more deliveries and avoids more 'cannot return' failures than standard shortest-path or replanning approaches.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The wind-error bound ε_W in Eq. 1 is never estimated or validated, and the paper admits gusts are unmodeled; EWR's safety guarantee hangs on this unvalidated bound.","rationale":"I read the central claim as an empirical comparison claim plus a safety-design claim. The empirical comparison is plausible given the reported success rates, but it is not reproducible without the missing formulas for ε_W, L_E, and the power model. The load-bearing weakness is not the comparison itself but the safety layer: Eq. 1's ε_W is the linchpin for every margin and feasibility check. The reader identified the same assumption, and the limitations section's admission about gusts supports the concern. No verdict adjustment is needed: the existing CONDITIONAL verdict already captures this, and acceptance should require the authors to provide explicit definitions of ε_W and L_E together with a validation trace against the replayed wind logs.","tokens_in":8478,"tokens_out":3675,"duration_ms":40862,"concrete_test":"Run the ASOS log-replay experiment with logging enabled: after each decision step t and for each relevant edge e, record the realized error norm ∥W(t,e) − Ŵ(t−τ_e,e)∥ and compare it with the ε_W(t,e) value used by EWR. Report the fraction of steps where the bound is violated, and whether any EWR return failure or negative M_min occurs in those violation episodes. If violations are non-negligible or coincide with failures, the guarantee is unsupported and the margin must be re-estimated; if violations are absent in all 100 rounds, the concern does not land.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central safety mechanism is Eq. 1's bounded-error model ∥ΔW_t(e)∥ ≤ ε_W(t,e) and Eq. 9's Lipschitz propagation ΔE_W = L_E(e) ε_W(t,e). EWR's adaptive margin, reachability filter, and return-feasibility checks are conservative only if ε_W is a true upper bound on realized wind-estimation error. The paper never gives the estimator for ε_W, never specifies L_E or the power terms in Eq. 5, and never validates either bound against the replayed ATS/ASOS wind logs. The conclusion explicitly states that 'unforeseen sudden strong wind gusts' are not modeled; if a gust pushes the error past ε_W, the claimed return-feasibility safety can fail exactly in the regime the method is designed for. The empirical success-rate comparison may still hold, but the central safety claim—and the meaningfulness of M_min in Eq. 12—is only as strong as this unvalidated bound.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes Energy-Aware Wind-Resilient Routing (EWR) for truck-assisted multi-UAV delivery. The environment is modeled as a time-dependent directed energy graph whose edge costs are computed from delayed noisy wind estimates, a bounded-error model, an adaptive safety margin, and a crosswind reachability filter. EWR replans online while checking that each UAV can safely return to the truck or depot. Experiments on synthetic graphs with replayed ATS and ASOS wind logs compare EWR against six baselines, reporting higher mission success rates, lower return failure rates, positive minimum energy margins, and sub-160 ms planning times.","tokens_in":8715,"tokens_out":5711,"duration_ms":57410,"significance":"If the safety margins and uncertainty bounds are fully specified and validated, the framework would be a useful contribution: it directly targets energy-feasible routing under wind uncertainty, a real operational concern, and it evaluates against multiple baselines with two wind-data sources. The paper is honest about its limitations, including unmodeled gusts and battery aging. The main empirical direction—that uncertainty-aware replanning with return-feasibility checks improves success rates—is plausible and worth publishing once the missing model definitions and validation are supplied.","major_comments":[{"comment":"The bounded-error model in Eq. (1) and the Lipschitz energy propagation in Eq. (9) are load-bearing but unspecified. The paper does not define how the confidence radius epsilon_W(t,e) is estimated from sensor data, how the Lipschitz bound L_E(e) is computed, or what functional forms P_parallel(t,e) and P_perp(t,e) take in Eq. (5). As a result, the conservative uncertainty margin E_safe + Delta E_W and the return-feasibility guarantee cannot be reproduced or tested. The conclusion's admission that \"unforeseen sudden strong wind gusts\" are not modeled further means that the guarantee holds only if the realized wind error stays inside epsilon_W, which is never validated against the replayed ASOS logs. Please specify these quantities and validate the bound empirically, or weaken the safety claims accordingly.","section":"III-B, Eq. (1), Eq. (9), Eq. (5)"},{"comment":"The reported minimum energy margin M_min is defined by subtracting exactly the quantities that EWR enforces as constraints: E_safe(t,e_t) and Delta E_W(t,e_t). Consequently, a positive M_min is in part a bookkeeping consequence of the routing rule rather than an independent safety certificate. For baselines it is unclear whether E_safe and Delta E_W are set to zero or to the same values; without this, the comparison in Fig. 4(b) may be biased. Please report a raw minimum margin (E_rem - E_t - E_ret without EWR's own safety terms) and state the baseline convention.","section":"V-B, Eq. (12), Fig. 4(b)"},{"comment":"Figure 3(a) reports mission success rates without error bars even though each setting uses 100 trials. Table III provides standard deviations for the real-wind-log setting, but the synthetic-setting figure does not; without variance information, the claimed improvements over Online-Replan and D* Lite cannot be assessed for statistical significance. Add error bars or confidence intervals to the figure.","section":"V-C, Fig. 3(a)"},{"comment":"The experimental setup does not describe how the wind logs are mapped to graph edges, what values of epsilon_W are used in simulation, or how the \"wind discretizations\" mentioned in the conclusion are varied. Section V-A only says the logs are streamed to update edge costs, and no experiment in the paper varies a discretization. Provide these implementation details or remove the claim about wind discretizations.","section":"V-A, VI"}],"minor_comments":[{"comment":"The columns for edges and planning time are concatenated in Table II (e.g., \"1863.8\" reads as 186 edges and 3.8 ms). Add a column separator or reformat the table.","section":"Table II"},{"comment":"The notation for airspeed is inconsistent: Eq. (4) uses V_A while Eq. (5) uses hat V_G(t,e); the definitions in Section III-A should state the relationship between these quantities.","section":"III-B"},{"comment":"The description of the \"public ATS-based truck-UAV delivery dataset\" (Ref. [26]) is minimal; specify how the 100 rounds are generated, how customers and UAVs are sampled, and how task allocation is performed before EWR routing.","section":"V-A"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper does one thing well: it wraps D* Lite-style replanning in a wind-aware energy graph with return-feasibility constraints and uncertainty margins, and the simulation evidence, including the ASOS log replay, supports the qualitative claim that EWR cuts return failures. The numbers in Table III are clearly separated, and the integration itself—delayed noisy wind estimates plus an uncertainty penalty plus a phase-dependent adaptive margin—is new as a package, even though each component exists in the literature.\n\nThe soft spots are real but not fatal. The biggest is the unvalidated error bound. Eq. 1 assumes ||ΔW|| ≤ ε_W, and Eq. 9 propagates that via an unspecified Lipschitz constant L_E; neither ε_W nor L_E nor the power terms P∥ and P⊥ in Eq. 5 is ever defined or validated against the replayed wind logs. That means the safety margin and the M_min values cannot be reproduced from the paper alone. And since M_min (Eq. 12) subtracts exactly the E_safe and ΔE_W terms that EWR enforces, a positive M_min for EWR is in part by construction; the objective evidence is the success and return-failure rates, which are fine. The conclusion also concedes that sudden wind gusts aren't modeled, so the return-feasibility guarantee is only as strong as the assumed bound. That's a genuine limitation, but the paper owns it, and the empirical trend is still believable.\n\nMinor issues: Table II's edges and planning-time columns run together (e.g., '1863.8' should be '186' and '3.8'), and Fig. 3(a) lacks error bars. Both fixable.\n\nI'd send this out for serious review. The claim is testable, the setup is sensible, and the missing parameters are a matter of completeness, not a wrong central idea. A good referee can push for the model details and a validation of ε_W on the actual logs. The paper would then be a solid applied addition to the truck-drone routing literature.","headline":"A solid engineering integration of wind-aware replanning with return-feasibility checks, but the central safety bound is unvalidated and the reported margin is partly by construction.","tokens_in":9224,"tokens_out":2878,"would_cite":false,"duration_ms":28662,"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":"This paper proposes an online routing framework that keeps delivery drones energy-safe under uncertain wind by building conservative uncertainty margins into every edge cost and return-feasibility check.","keywords":["truck-UAV delivery","wind uncertainty","energy-aware routing","return feasibility","risk-sensitive planning","time-dependent energy graph","online replanning","UAV energy safety"],"falsifier":"Replay the same weather-station wind log through EWR while recording, at each decision step, the actual deviation between the true wind and the delayed estimate; if that deviation exceeds the assumed $\\epsilon_W(t,e)$ on any step that is followed by a return failure or a negative minimum margin, the conservative bound is not conservative. A cleaner version of the experiment inserts a short gust event exceeding $\\epsilon_W$ into the log and checks whether EWR's return-failure rate rises to the level of Online-Replan or D* Lite.","tokens_in":8283,"feed_emoji":"🚁","tokens_out":7982,"duration_ms":78855,"temperature":0.7,"pith_summary":"This paper argues that routing decisions for truck-assisted multi-UAV delivery should be made against a time-varying energy graph whose edge costs include not only the current wind estimate but a conservative margin for how wrong that estimate might be. Its central proposal, EWR, replans each UAV's forward path online while checking at every step whether the UAV can still return to the truck or depot with the margin intact. The authors test this in simulated delivery graphs with wind data replayed from a public truck-UAV delivery dataset and from a real weather station, and they report higher mission success rates and lower wind-induced return-failure rates than shortest-path, energy-aware, and online-replanning baselines. The result matters because low-battery, high-wind conditions are precisely where static energy models underestimate headwind, crosswind, and return-feasibility risk.","feed_headline":"Wind-margin routing lifts drone delivery success to 86%","feed_subtitle":"Adding margins for wind error to every energy check cuts drone return failures to below 3 percent.","key_machinery":"The central object is a time-dependent directed energy graph $G_t=(V,E,c_t)$ whose edge cost is $c_t(e)=E_t(e)+\\lambda U_t(e)$: predicted traversal energy plus a wind-uncertainty penalty. Three components carry the argument: the bounded-error wind model $W(t,e)=\\hat{W}_t(e)+\\Delta W_t(e)$ with $\\|\\Delta W_t(e)\\|\\le \\epsilon_W(t,e)$; the conservative reachability filter that rejects near-critical edges when the crosswind estimate plus its error bound exceeds a lateral control margin; and the adaptive safety margin $E_{\\mathrm{safe}}(t,e)$ built from an emergency reserve, wind and battery state-of-charge uncertainty, and mission phase, with wind-induced energy uncertainty upper bounded by the Lipschitz product $L_E(e)\\epsilon_W(t,e)$. The recursive use of these bounds makes every online replan an energy-feasibility check rather than merely a shortest-path update.","core_discovery":"Under delayed, noisy wind observations, a UAV's safe-return guarantee cannot come from the current wind estimate alone. EWR's claim is that routing with an explicit uncertainty budget—a bounded confidence radius on wind error, a conservative power-model Lipschitz bound, and an adaptive safety margin—keeps the minimum residual energy margin positive throughout the flight, whereas replanning methods that trust the latest estimate let the margin go negative and end in return failures. In the paper's experiments this shows up as roughly 92% mission success at full battery and above 75% at half battery, an 86.5% success rate under real weather-station wind replay with a 2.8% return-failure rate, and per-step planning times under 160 ms at the largest tested graph size.","pith_inferences":["If the margin framework is sound, it can be layered on top of any existing replanner as a safety wrapper: the return-feasibility filter checks a candidate path before execution, so it could guard non-margin baselines without redesigning them.","A natural next step the paper does not take is to make the confidence radius $\\epsilon_W(t,e)$ an online-estimated quantity, for example from the residual between forecast and observed wind, which would turn a fixed margin into an adaptive one.","Combining the margin approach with gust-prediction models would directly address the paper's stated limitation that sudden wind gusts are not modeled; the current worst-case bound only covers errors inside $\\epsilon_W$.","The framework's metric $M_{\\min}$ could double as a real-time alarm: a controller watching it cross zero has an immediate trigger for safe abort, which fits the paper's distinction between aborts and return failures."],"forward_implications":["Adding an uncertainty margin to edge costs and an explicit return-feasibility check at each replan step is what separates EWR from reactive replanners in the paper's tests; without the margin, Online-Replan and D* Lite let the minimum energy margin go negative under the lower battery budget.","The reported per-step planning times below 160 ms at 352 nodes and 3486 edges suggest that the margin-carrying graph search remains fast enough for online use in medium-scale truck-UAV delivery.","A positive minimum energy margin, not just a higher success count, is the operational claim: EWR reports positive $M_{\\min}$ under both battery budgets, while most baselines report near-zero or negative margins.","On real weather-station wind logs, EWR reports an 86.5% success rate and a 2.8% return-failure rate at full battery, versus 58-78% success and 8-20% failure for the baselines, indicating that the benefit survives realistic wind variability."],"supporting_citations":[{"why":"Supplies the public truck-UAV delivery dataset whose time-indexed wind logs are streamed into the simulated graph to update edge costs.","marker":"[26]"},{"why":"Provides the public weather-station data retrieval interface used to obtain the real-world wind log for validation.","marker":"[29]"},{"why":"Provides the automated surface weather observations that serve as the real-world wind-log source.","marker":"[28]"},{"why":"Supplies the empirical multi-rotor power-consumption values used to set cruise power, battery capacity, and emergency reserve in the simulation.","marker":"[27]"},{"why":"Defines the energy-aware shortest-path routing baseline (Energy-SP) that EWR is compared against.","marker":"[15]"},{"why":"Defines the non-margin online replanning baseline that trusts the latest wind estimate without uncertainty margins.","marker":"[24]"},{"why":"Defines the incremental dynamic-graph replanning baseline (D* Lite) used as a classical comparison.","marker":"[25]"},{"why":"Defines the fixed-cost no-wind shortest-path baseline (SP-NoWind) used to show the cost of ignoring wind.","marker":"[23]"}],"fun_headline_variants":["Wind-uncertainty budget keeps drone energy positive","Energy-safe routing with wind-error margins lifts success","Uncertain wind? Route with a safety margin for energy","Wind-error budgets cut drone return failures to under 3%","Adaptive wind-error margins keep UAV energy positive"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The safety margin only works if the true wind never differs from the delayed estimate by more than the confidence radius the planner assumes, and the paper does not show how that radius is estimated or validated, while it explicitly acknowledges that sudden gusts are not modeled—so an unexpected gust bigger than the assumed bound would break the return-feasibility guarantee precisely where it matters.","fun_headline_variants_meta":{"raw":{"variants":["Wind-uncertainty budget keeps drone energy positive","Energy-safe routing with wind-error margins lifts success","Uncertain wind? Route with a safety margin for energy","Wind-error budgets cut drone return failures to under 3%","Adaptive wind-error margins keep UAV energy positive"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000393,"raw_usage":{"total_tokens":2016,"prompt_tokens":852,"completion_tokens":1164,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":468,"completion_tokens_details":{"reasoning_tokens":1087}},"tokens_in":468,"tokens_out":1164,"duration_ms":10133,"temperature":1.0,"reasoning_tokens":1087,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T00:32:01.111760+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Replay the same weather-station wind log through EWR while recording, at each decision step, the actual deviation between the true wind and the delayed estimate; if that deviation exceeds the assumed $\\epsilon_W(t,e)$ on any step that is followed by a return failure or a negative minimum margin, the conservative bound is not conservative. A cleaner version of the experiment inserts a short gust event exceeding $\\epsilon_W$ into the log and checks whether EWR's return-failure rate rises to the level of Online-Replan or D* Lite.","supporting_citations":[{"cited_title":"Iowa Environmen- tal Mesonet: Download ASOS/AWOS/METAR Data,","cited_arxiv_id":null,"evidence_quote":"Provides the public weather-station data retrieval interface used to obtain the real-world wind log for validation."},{"cited_title":"1-Minute Page 1 Surface Weather Observations from the Automated Surface Observing System Network,","cited_arxiv_id":null,"evidence_quote":"Provides the automated surface weather observations that serve as the real-world wind-log source."},{"cited_title":"Empirical power consumption model for uavs,","cited_arxiv_id":null,"evidence_quote":"Supplies the empirical multi-rotor power-consumption values used to set cruise power, battery capacity, and emergency reserve in the simulation."},{"cited_title":"Shortest path planning for fleets of delivery drones: Evaluating the benefit of including a temporal dimension in the routing network,","cited_arxiv_id":null,"evidence_quote":"Defines the non-margin online replanning baseline that trusts the latest wind estimate without uncertainty margins."},{"cited_title":"Koenig and M","cited_arxiv_id":null,"evidence_quote":"Defines the incremental dynamic-graph replanning baseline (D* Lite) used as a classical comparison."},{"cited_title":"Comparative analysis of uav path planning algorithms for efficient navigation in urban 3d environ- ments,","cited_arxiv_id":null,"evidence_quote":"Defines the fixed-cost no-wind shortest-path baseline (SP-NoWind) used to show the cost of ignoring wind."}],"review_version":1}