{"id":"00e56f08-0de3-45fb-8810-e770443f26b0","arxiv_id":"1908.05905","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A drone-assisted on-demand routing protocol improves packet delivery ratio and reduces end-to-end delay in sparse urban vehicular networks, but its claimed superiority in dense traffic is contradicted by the paper's own results.","lead":"This paper proposes a routing protocol for connected cars in cities that uses drones as relay nodes when the vehicle network is too sparse to deliver messages. Simulations show higher message delivery and lower delay than two existing protocols in sparse conditions, and mixed results in dense conditions.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Connectivity gain rests on an unenforced UAV-coverage assumption: III.A promises a hovering UAV over every four road segments, but the simulation uses 16 MobiSim UAVs moving at 50–120 km/h, so the reported gain may be an artifact of random placement rather than the protocol.","rationale":"The reader's weakest assumption identifies the same load-bearing premise: continuous UAV coverage at the fragmentation points. My read strengthens this concern from a real-world feasibility worry to an internal consistency issue. Section III.A promises that at every moment at least one UAV hovers over an area of four road segments, but the simulation setup (Table II) gives UAVs a speed of 50–120 km/h and says their mobility is generated by MobiSim, so the simulation cannot be enforcing the hover requirement. Without a placement or coverage-control mechanism, the claimed connectivity extension rests on an assumption that the presented NS-2 experiments do not verify. The end-to-end delay contradiction at high densities identified by the reader is real but secondary; the coverage premise is more fundamental because it underpins both the PDR and the delay results in the sparse regimes where the paper claims its main advantage. The issue is addressable through targeted simulation instrumentation and a hover-vs-mobile comparison, so a conditional verdict with requests for those checks remains appropriate; the reader's CONDITIONAL verdict does not need to change.","tokens_in":8304,"tokens_out":5286,"duration_ms":51584,"concrete_test":"Re-run the NS-2 scenarios with 80–120 vehicles, instrumenting every RERR and every failed greedy forward to record (i) whether a UAV is within 300 m of the disconnection point and (ii) whether that UAV can also reach the next zone on some stored alternative path. If the fraction of events with no such UAV is non-negligible (e.g., >5–10%), the III.A coverage assumption is not realized by the simulation. A cleaner variant: compare the current MobiSim UAV trajectories at 50–120 km/h against a configuration in which UAVs hover in place over prescribed four-segment areas, as III.A assumes. If the PDR and delay curves differ materially, the reported results depend on unstated positioning rather than on the routing protocol.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section III.A states: \"we suppose that the network has a sufficient number of UAVs so that at each moment, at least one UAV hovers an area of four road segments.\" This is the load-bearing premise for the central claim. In a fragmented urban VANET, a UAV can only extend connectivity if it is physically within communication range (≈300 m) of the two disconnected ground segments at the instant a path is needed. The paper provides no mechanism that places or holds UAVs at such gaps: UAV mobility is generated by MobiSim with speeds of 50–120 km/h (Table II), which is not hovering, and the conclusion lists \"make the mobility of UAVs more controllable\" as future work. The simulation therefore does not verify the protocol under its own stated assumption; it instead appears to rely on random UAV positions providing the required bridge whenever a route breaks. If the coverage guarantee is not met, the RREQ/RREP discovery and the path-maintenance fallback to UAV hops have no connectivity to exploit, and the claimed PDR and delay advantages over RBVT-R and AGP would not follow.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a reactive routing protocol for urban VANETs in which UAVs participate in route discovery, path scoring, and path maintenance. Paths are scored using a multi-criteria formula that rewards vehicle count, penalizes delay, density imbalance, and the number of UAV hops, and the best path is selected from multiple discovered routes. When a route breaks, an alternative pre-discovered path is used, and UAVs are employed to bridge disconnected ground clusters. The protocol is evaluated in NS-2 with VanetMobiSim/MobiSim mobility against RBVT-R and AGP, reporting packet delivery ratio, end-to-end delay, hop count, and routing overhead. The abstract claims reduced end-to-end delay and high delivery ratio, supporting the conclusion that heterogeneous vehicle-UAV communication extends network connectivity.","tokens_in":8542,"tokens_out":3664,"duration_ms":35757,"significance":"If the claimed results hold, the paper would make a moderate contribution to UAV-assisted VANET routing by combining zone-density-aware path scoring with UAV relaying and pre-computed alternative paths. The comparison against RBVT-R and AGP is appropriate, and the scoring rule in Eq. (3) is a constructive, parameter-free heuristic rather than a fit to the reported outcomes. However, the central claim rests entirely on simulation evidence that conflicts with the paper's own coverage assumption and with its internal delay statements; no statistical support is provided. The idea is plausible and worth pursuing, but the current manuscript does not establish the load-bearing connectivity claim.","major_comments":[{"comment":"The coverage assumption in Section III.A that the network has a sufficient number of UAVs so that at each moment at least one UAV hovers over an area of four road segments is never enforced or verified in the simulation. Table II specifies only 16 UAVs moving at 50–120 km/h, and the conclusion lists making UAV mobility more controllable as future work. Since the claimed connectivity gain depends on a UAV being within communication range of two disconnected ground segments at the instant a path is needed, the simulation results may reflect random UAV placement rather than the routing protocol itself. The authors should either implement a UAV control/placement model that realizes the hovering assumption or evaluate the protocol under conditions where the assumption is relaxed.","section":"III.A, Table II"},{"comment":"The results text in Section IV.C explicitly states that for more than 140 vehicles, RBVT-R outperforms the proposed approach in end-to-end delay, and that UAV-composed routes are unstable due to high mobility. The Conclusion, however, claims the protocol outperforms existing routing in terms of end-to-end delay 'especially the high densities,' and the abstract states an unqualified 'reduced end-to-end delay.' These statements are internally contradictory and overstate the findings. The claims must be restricted to the density range where the data support them.","section":"IV.C, Fig. 7(b), Conclusion"},{"comment":"The evaluation reports 15 simulation repeats per scenario, but no error bars, confidence intervals, or statistical tests are presented for any metric. This is particularly important because the end-to-end-delay advantage appears only below roughly 140 vehicles, and the crossover is a central point of the paper's claims. Without statistical evidence, the differences shown in Fig. 7 could be within simulation noise.","section":"IV.A, IV.C"},{"comment":"The score in Eq. (3) depends on a Delay value carried in the RREQ, but the manuscript does not specify how this delay is estimated during route discovery, which nodes update it, or how the destination obtains the values shown in Table I. The RREQ description says UAVs add only their IDs to the transited-zones list, so it is unclear how UAV-involved paths accumulate the delay metric. Without a defined delay estimator, the scoring rule is not reproducible.","section":"III.C, Eq. (3)"}],"minor_comments":[{"comment":"The definition of average number of hops, 'the number of successfully delivered data packets divided by the total number of hops,' is the inverse of the intended quantity; the figure axes and the surrounding text indicate the metric is hops per delivered packet. Please correct the definition.","section":"IV.B"},{"comment":"The phrase 'RBVT-R performs well than AGP' should read 'RBVT-R performs better than AGP.'","section":"IV.C"},{"comment":"The word 'undiserable' should be 'undesirable.'","section":"III.C"},{"comment":"Reference [12] is cited as a 2016 publication for VanetMobiSim; please verify the correct year and version, as the mobility generator is widely known from earlier releases.","section":"II"},{"comment":"The simulation map is described only as 4000 m × 4000 m with 9 intersections; the road lengths, block sizes, and segment structure are not given, which makes it difficult to interpret the meaning of the zone sizes and vehicle densities.","section":"IV.A"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is closely related to the authors' earlier protocol in reference [1]; the revision should clarify the concrete differences and the incremental contribution of the new scoring, maintenance, and UAV-cooperation mechanisms relative to that prior work."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe thing to know: this paper has a sensible reactive routing protocol that combines zone-density scoring with UAV relays, but the simulation evidence is weaker than the abstract claims, and the load-bearing assumption about UAV coverage is never actually enforced.\n\nWhat's new: the specific combination — fixed zones, standard-deviation-based density balance, alternative paths stored in the data header, and UAVs as a final bridge when the ground network fragments — doesn't appear in the cited prior work. That is a legitimate, incremental extension. The scoring equation is explicit and not fitted to the outcomes, so there's no circularity. The maintenance idea, using pre-discovered paths before falling back to drones, is a reasonable design choice.\n\nThe soft spots, in rough order. First, Section III.A assumes enough UAVs that one hovers over every four road segments, but the simulation uses 16 UAVs moving at 50–120 km/h with no mechanism that puts them at fragmentation points. That means the improved PDR might come from fortuitous random placement rather than the protocol's discovery and maintenance logic. The authors even list 'make the mobility of UAVs more controllable' as future work, which tells you they know the assumption isn't implemented. Second, the conclusion claims the protocol outperforms on delay 'especially the high densities,' which is directly contradicted by Fig. 7b and their own results section, where RBVT-R wins above 140 vehicles. That needs fixing — it's a simple overstatement, but in a short paper it undermines trust. Third, there are no error bars or statistical tests despite 15 repeats, and no code or data release. Fourth, the closest baseline is the authors' own 2017 UAV-assisted protocol, which is cited but not compared against.\n\nThe math is simple and mostly fine; the scoring equation is arbitrary but not derived by fitting, so no circularity. The citation pattern is clean.\n\nThis paper is for people working on UAV-assisted VANET routing. It deserves a serious referee only if the coverage assumption is addressed and the delay conclusion corrected. As is, I'd treat it as a conditional paper with a promising core. A journal editor could reasonably send it to review, because the flaws are addressable and the protocol idea is not silly — but I'd expect a major revision, not a quick accept.\n\nRecommendation: engage with it, but make the authors prove the UAV placement and fix the internal contradiction before relying on the results.","headline":"Plausible protocol, unenforced UAV-coverage assumption, and a conclusion that contradicts its own delay plot.","tokens_in":9025,"tokens_out":3627,"would_cite":false,"duration_ms":35011,"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":"A new reactive routing protocol that makes UAVs full members of discovered paths and uses them as bridges for fragmented ground segments can extend connectivity in sparse urban VANETs, reducing end-to-end delay and raising packet delivery…","keywords":["VANET routing","UAV-assisted routing","reactive routing","urban vehicular networks","path scoring","network connectivity","packet delivery ratio","end-to-end delay"],"falsifier":"Run the same protocol in a simulator where drone coverage is not guaranteed over every four road segments, or where building obstacles attenuate air-to-ground links, and measure whether the packet-delivery-ratio advantage over ground-only routing vanishes below a threshold such as fewer than the assumed 16 UAVs.","tokens_in":8114,"feed_emoji":"🚁","tokens_out":4488,"duration_ms":41540,"temperature":0.7,"pith_summary":"The paper tries to establish that a reactive routing protocol for urban vehicular networks can keep delivery reliable in sparse traffic by recruiting drones as full members of routing paths and as bridges when ground connectivity fragments. The proposed scheme scores every discovered path by vehicle count, delay, density balance, and number of UAV hops, then selects the best path and keeps alternative paths for instant failover. In packet-level simulation against two existing urban VANET protocols, it reports higher packet delivery ratio and lower end-to-end delay, especially when the vehicle count is low. If true, the significance is that heterogeneous vehicle-drone communication is a practical way to extend a VANET's usable range without waiting for dense traffic.","feed_headline":"Drones as relay hops keep sparse city VANETs connected","feed_subtitle":"A reactive score picks balanced vehicle-dense paths and switches to UAV bridges when ground links break.","key_machinery":"The central object is a routing path represented as a sequence of fixed zone IDs, where each zone is roughly the size of the vehicle communication range (about 300 m). During route discovery, a RREQ packet accumulates each transited zone ID and the vehicle count in that zone, so the destination can compute the average zone density and the standard deviation of zone densities for every discovered path. The selection score combines those computed metrics with the raw delay and the number of UAV hops, and the winning path is returned to the source greedily. The maintenance machinery stores all discovered paths in the data-packet header, so an intermediate node can switch to an alternative path without re-running discovery, and when no ground alternative exists, a UAV is designated as the next forwarder to bridge disconnected clusters.","core_discovery":"The paper claims that a heterogeneous network in which UAVs are full members of on-demand routing paths, rather than mere data mules, keeps urban vehicular networks connected when ground traffic is sparse. The destination scores each candidate path by $$\\mathrm{Score}=\\frac{\\mathrm{NB_{vehicles}}}{\\mathrm{Delay}}\\times \\frac{1}{1+\\mathrm{Sdeviation}+\\mathrm{Hops_{UAV}}}$$ where NB_vehicles is the number of vehicles along the path, Delay is the RREQ traversal time, Sdeviation is the standard deviation of per-zone vehicle counts, and Hops_UAV is the number of UAV hops in the path. Paths with more vehicles, lower delay, balanced vehicle distribution, and fewer UAV hops are preferred; all discovered paths are stored in the data-packet header, and when the selected path breaks, the current forwarder switches to an alternative path, or calls on a UAV to bridge the gap. In NS-2 simulations over a 4 km x 4 km grid with up to 200 vehicles and 16 UAVs, the protocol reports higher packet delivery ratio and lower end-to-end delay than RBVT-R and AGP at low vehicle densities, though at high densities RBVT-R achieves lower delay.","pith_inferences":["The protocol assumes blanket drone coverage over every group of four road segments; a more economical design would reposition UAVs toward predicted disconnection points rather than hovering everywhere.","The score weights vehicle density, delay, density balance, and UAV hops equally; changing those relative weights could trade path stability against delay and could be tuned separately for highway or rural settings.","If air-to-ground links suffer the same building blockage as ground-to-ground links, the bridge advantage shrinks, so a simulation varying building height and link budget would test the core connectivity claim.","The zone-density scoring idea could be reused in non-urban settings, which the paper itself names as a possible extension."],"forward_implications":["In sparse urban traffic, routing paths need not be confined to ground vehicles: UAV hops extend connectivity and reduce the average number of hops.","Storing all discovered paths in the data-packet header lets intermediate nodes switch paths without a new route discovery as long as at least one alternative remains.","Density balance, not just shortest distance or fewest intersections, is a workable route-selection criterion for city VANETs.","At high vehicle densities, ground-only reactive routing can match or beat the UAV-assisted approach in end-to-end delay, because UAV mobility triggers route rediscovery.","When alternatives are exhausted, the protocol falls back to a route-error message and a fresh discovery, so failures degrade to standard reactive behavior."],"supporting_citations":[{"why":"Baseline reactive protocol that selects shortest-distance paths; the paper compares its density-aware scoring and recovery strategy against it.","marker":"[4]"},{"why":"Baseline anchor-geography protocol with mobility prediction; compared for packet delivery ratio, delay, hops, and overhead.","marker":"[7]"},{"why":"Prior UAV-assisted disruption-tolerant routing where UAVs only carry packets and do not use geographic positions, motivating UAVs as full path members.","marker":"[10]"},{"why":"The authors' earlier intelligent UAV-assisted routing protocol for urban VANETs, extended here with multi-path on-demand scoring and maintenance.","marker":"[1]"},{"why":"Generates the vehicular mobility traces used in the simulation experiments.","marker":"[12]"},{"why":"Generates the UAV mobility traces used in the simulation experiments.","marker":"[13]"}],"fun_headline_variants":["UAVs join VANET routing paths to bridge sparse city gaps","Drones as full routing hops keep sparse VANETs connected","Scoring paths with vehicle balance and UAV hops boosts delivery","Heterogeneous routing: UAVs as active members, not just mules"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The scheme needs a sufficient number of UAVs that each area of four road segments always has at least one UAV overhead, and it assumes drone-to-vehicle wireless links are long-range and unaffected by buildings; without those two conditions the connectivity gain disappears.","fun_headline_variants_meta":{"raw":{"variants":["UAVs join VANET routing paths to bridge sparse city gaps","Drones as full routing hops keep sparse VANETs connected","Scoring paths with vehicle balance and UAV hops boosts delivery","Heterogeneous routing: UAVs as active members, not just mules"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000176,"raw_usage":{"total_tokens":1308,"prompt_tokens":980,"completion_tokens":328,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":596,"completion_tokens_details":{"reasoning_tokens":254}},"tokens_in":596,"tokens_out":328,"duration_ms":4197,"temperature":1.0,"reasoning_tokens":254,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:01:04.020860+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same protocol in a simulator where drone coverage is not guaranteed over every four road segments, or where building obstacles attenuate air-to-ground links, and measure whether the packet-delivery-ratio advantage over ground-only routing vanishes below a threshold such as fewer than the assumed 16 UAVs.","supporting_citations":[{"cited_title":"V ANET routing on city roads using real-time vehicular traffic information,","cited_arxiv_id":null,"evidence_quote":"Baseline reactive protocol that selects shortest-distance paths; the paper compares its density-aware scoring and recovery strategy against it."},{"cited_title":"AGP: an anchor-geography based routing protocol with mobility prediction for vanet in city scenarios,","cited_arxiv_id":null,"evidence_quote":"Baseline anchor-geography protocol with mobility prediction; compared for packet delivery ratio, delay, hops, and overhead."},{"cited_title":"UA V assisted disruption tolerant routing,","cited_arxiv_id":null,"evidence_quote":"Prior UAV-assisted disruption-tolerant routing where UAVs only carry packets and do not use geographic positions, motivating UAVs as full path members."},{"cited_title":"Intelligent UA V-Assisted Routing Protocol for Urban V ANETs,","cited_arxiv_id":null,"evidence_quote":"The authors' earlier intelligent UAV-assisted routing protocol for urban VANETs, extended here with multi-path on-demand scoring and maintenance."},{"cited_title":"VanetMobiSim: generating realistic mobility patterns for V ANETs,","cited_arxiv_id":null,"evidence_quote":"Generates the vehicular mobility traces used in the simulation experiments."},{"cited_title":"Mobi-sim: A simulation en- vironment for mobile wireless sensors network,","cited_arxiv_id":null,"evidence_quote":"Generates the UAV mobility traces used in the simulation experiments."}],"review_version":1}