{"id":"718b9316-5d49-4416-bc66-ffcfc4004bed","arxiv_id":"2505.03331","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A 9-gram integrated multihole pressure probe with a cone tip and 1.2 mm hole spacing estimates airspeed (MAE 0.44 m/s) and flow angles (MAE 0.16°) from 3 to 27 m/s, with outdoor flight validation near stall.","lead":"What did this paper find or do: A new lightweight 9-gram airflow probe for small drones measures airspeed, angle of attack, and sideslip at speeds as low as 3 m/s, using five differential pressure sensors packed into one printed part. Why read it: It may make stall protection practical for lightweight UAVs and is released as open hardware with data.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Calibration test set includes linearly interpolated points, so the reported 0.44 m/s and 0.16° MAEs may be optimistically biased; re-evaluate on measured points only.","rationale":"The reader's weakest_assumption identifies the same load-bearing concern: augmented interpolated points in the test set can deflate the reported calibration error. I agree that this is the most serious threat to the central quantitative claim, because the abstract and conclusion rely directly on the 0.44 m/s and 0.16° wind-tunnel MAEs. The paper does provide real supporting evidence: open-source hardware, a concrete wind-tunnel dataset, and outdoor flights that demonstrate the concept. Those elements justify a conditional, not a rejection, verdict. However, the interpolation issue is internal to the evaluation protocol and is testable by recomputing the error on measured points only. The outdoor validation also uses non-ground-truth references and reports larger errors, so it cannot independently rescue the headline numbers. I therefore keep the reader's CONDITIONAL verdict unchanged, pending the re-evaluation described in the concrete test.","tokens_in":8407,"tokens_out":2588,"duration_ms":25499,"concrete_test":"Re-run the calibration pipeline of Section III-B with all interpolated points assigned to the training set (or removed from the test set), then evaluate on the 30% held-out original measured points only. Report MAE/RMSE for airspeed, AoA, and AoS, ideally with 5-fold stratified cross-validation over the 17 angle configurations and 9 speeds. If the measured-only MAE remains close to 0.44 m/s and 0.16°, the interpolation concern is resolved; if it rises substantially (e.g., airspeed MAE >0.7 m/s or angular MAE >0.5°), the headline accuracy should be revised and the outdoor-validation caveats strengthened.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The headline accuracy claim rests on a train/test split performed after augmenting the regression dataset with linearly interpolated artificial points (Section III-B, Fig. 7A). Because each artificial point is a convex combination of nearby measured points, and the degree-3 polynomial is fit on a shuffled 70% subset, any augmented point landing in the 30% test set will be nearly exactly predictable from its parent measured points if those parents are in the training set. The reported test MAE (0.44 m/s, 0.163°, 0.156°) therefore does not measure generalization to unseen flow conditions; it partly measures how well the model reproduces linear interpolants. The manuscript does not state that augmented points are excluded from test evaluation or that the reported errors are computed on original measured points only. This is load-bearing because the abstract and conclusion present these numbers as the sensor's accuracy, and the outdoor validation errors (1.65 m/s, 3.20°, 5.87°) are larger and based on non-ground-truth references (Section IV). If the true wind-tunnel MAE on measured-only test points is materially higher, the central claim of 'accurate estimations' at low speed is weakened.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a 9 g multihole pressure probe with five differential pressure sensors integrated into a single 3D-printed structure, avoiding external tubing. The authors compare eight hardware variants (two tip shapes and four hole spacings) in an open wind tunnel, select a conical-tip, 1.2 mm spacing design, calibrate it with a degree-3 multivariate polynomial regression over 3-27 m/s airspeed and +/-35 degrees angle of attack and sideslip, and validate the sensor in outdoor flights on a fixed-wing UAV. The headline results are a mean absolute error of 0.44 m/s for airspeed and 0.16 degrees for angles in the wind-tunnel test set, with larger errors in outdoor flight, and the authors argue the sensor fills a gap for lightweight UAVs flying near the stall regime.","tokens_in":8615,"tokens_out":5252,"duration_ms":55405,"significance":"If the reported wind-tunnel accuracy is reliable, this is a useful contribution: a lightweight, open-source multihole probe that operates at low airspeeds would address a real gap, since existing miniature probes typically require airspeeds above 20 m/s. The hardware design study (eight variants), the public release of the design and data repository, and the outdoor flight validation are all strengths. However, the central accuracy claim is weakened by the calibration protocol: linearly interpolated artificial points are added before the train/test split, which can artificially lower the reported test MAE. The outdoor validation also relies on non-ground-truth references. The contribution is promising, but the evidence for the headline accuracy numbers needs revision.","major_comments":[{"comment":"The reported test MAE (0.44 m/s, 0.163 degrees, 0.156 degrees) is computed after adding linearly interpolated artificial points to the regression dataset before the 70/30 shuffled split. Because each artificial point is a convex combination of nearby measured points, an artificial point assigned to the test set can be predicted almost exactly from its parent measured points if those parents are in the training set. The test error therefore partly measures how well the model reproduces linearly interpolated targets rather than generalization to unseen flow conditions. Please recompute and report the MAE and RMSE using only the original measured points for testing, and state explicitly whether any augmented points are excluded from the test subset.","section":"Section III-B, Fig. 7A"},{"comment":"The manuscript does not specify the unit of the regression dataset used for the 70/30 split: whether each sample is the averaged static portion of a 20 s measurement for one (speed, AoA, AoS) configuration, or whether the 504,900 individual filtered time samples are used. If individual time samples from the same static segment appear in both training and test sets, the split is not independent and the reported errors are biased downward regardless of the interpolation issue. Please clarify the number and granularity of samples entering the regression and, if time samples are used, perform the split at the configuration level.","section":"Section III-B, Fig. 7B"},{"comment":"The outdoor validation references are not ground truth. The AoA and AoS estimates are derived from the Pixhawk autopilot's body-frame velocity estimate, which combines IMU, GPS, and pitot data and assumes zero wind. The statement in Section IV that the results 'demonstrate that our sensor and its calibration perform well in an unseen, applied outdoor setting and did not overfit' is stronger than the evidence supports. Please either validate against an independent reference (for example, a calibrated wind vane, motion capture, or a known maneuver) or explicitly qualify the outdoor results as relative comparisons rather than accuracy validation.","section":"Section IV, Eqs. (2)-(3)"},{"comment":"The conclusion states that the probe 'achieves an RMSE of 0.22 degrees for angle estimation and 3.6% for speed' and compares these numbers with literature, but no RMSE values or their computation basis are reported in Section III. The text reports only MAE for the test set. Please provide the RMSE values together with the same test-set definition used for the MAE, or remove the unsupported comparison.","section":"Section V"}],"minor_comments":[{"comment":"The term 'UA V' appears with an odd space in the abstract and introduction; use 'UAV' consistently.","section":"Throughout"},{"comment":"The sensor is described as having 13 holes in Section II-A but as a 'nine-hole' multihole pressure probe in the conclusion; please reconcile the terminology for the number of holes.","section":"Section II-A and Section V"},{"comment":"The y-axis of the elbow plot is not labeled; it is unclear whether the plotted quantity is MAE, RMSE, or another error metric. Please label the axis and state the metric in the caption.","section":"Section III-B, Fig. 7C"},{"comment":"The notation in Table I (avg/std reductions over five dimensions) is difficult to parse. Consider presenting the four metrics as explicit equations or a clearer table with row and column headers.","section":"Table I"},{"comment":"References [2] and [3] cite Fandom and Wikiwand pages rather than archival accident reports or peer-reviewed sources; please replace them with authoritative references.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The interpolation-before-split issue is the main technical barrier. If the authors recompute the metrics on measured-only test points and the accuracy remains competitive, the paper would be suitable for publication. I also recommend requesting clarification of the sample granularity for the regression split, since temporal leakage would be a separate and serious problem."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know about arXiv:2505.03331. First, the hardware is genuinely new and practical: integrating the PCB into the printed probe with differential sensors on paired holes removes tubing and halves the sensor count, and the whole thing is 9 g and released to the public domain. That is a real advance for lightweight UAV stall protection, because existing MPPs are either heavier or need airspeeds above 20 m/s. Second, the headline wind-tunnel accuracy numbers are probably optimistic. The calibration holdout in Section III-B includes artificially linearly interpolated points added before the train/test split, and the paper never states that the test evaluation is restricted to original measured points. Since each interpolated point is a convex combination of nearby measured points, and the polynomial is fit on a shuffled 70% subset, any interpolated point landing in the test set will be nearly predictable from parents likely sitting in the training set. The stress-test concern lands: the 0.44 m/s and 0.16° MAEs partly measure reproduction of interpolants, not generalization to unseen flow conditions. That is a load-bearing flaw in the abstract's central claim, but not in the design itself.\n\nWhat the paper does well: the eight-design comparison of tip shape and hole spacing is systematic and provides genuinely new data for low-speed MPP design. Conical tips win on resolution and noise; hole spacing barely matters. The q-scaling normalization for differential sensors is a reasonable preprocessing step, and the authors openly acknowledge that the outdoor comparisons are not ground truth, which is more honest than many papers in this area.\n\nThe other soft spots are in proportion. The calibration grid is sparse—17 angle configurations over ±35°—so the polynomial is doing a fair amount of interpolation over the angular range; that is not fatal for a stall probe but worth stating. The outdoor validation MAEs (1.65 m/s, 3.20°, 5.87°) are substantially worse than the wind-tunnel numbers, so the sentence in Section IV that the sensor \"did not overfit\" is stronger than the evidence supports. The fix is straightforward: re-estimate the test MAE on measured points only, report confidence intervals, and present the outdoor errors with the acknowledged caveats in the same place as the wind-tunnel numbers.\n\nOverall: the engineering is solid, the flaws are in evaluation reporting rather than in the core idea. A serious referee would be worth the time. My recommendation is to send it to peer review, and to require the measured-only test recalculation before acceptance.","headline":"A genuinely lightweight 9 g multihole probe with a systematic design study; the headline wind-tunnel MAEs are probably optimistic because the test set includes interpolated points, but the engineering contribution is solid and deserves careful peer review.","tokens_in":9185,"tokens_out":2014,"would_cite":true,"duration_ms":21252,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A 9 g multihole probe estimates airspeed, angle of attack, and sideslip from 3 to 27 m/s and ±35° using a third-degree polynomial calibration.","keywords":["multihole pressure probe","airspeed estimation","angle of attack","angle of sideslip","stall detection","lightweight UAV","differential pressure sensors","polynomial regression"],"falsifier":"Recalibrate all three polynomials on only the 17 measured wind-tunnel configurations, without the interpolated artificial points, then evaluate on newly measured wind-tunnel points across the same 3-27 m/s and ±35° grid; if the mean absolute error remains near 0.44 m/s and 0.16°, the central claim stands, and if it rises substantially, the headline accuracy is an artifact of the interpolation.","tokens_in":8177,"feed_emoji":"💨","tokens_out":7103,"duration_ms":65896,"temperature":0.7,"pith_summary":"This paper tries to close a gap in small-drone instrumentation: existing multihole pressure probes are either heavier than 30 g or need airspeeds above 20 m/s, while lightweight fixed-wing UAVs often fly below 15 m/s and are most at risk near stall. The authors propose a 9 g single-component probe that measures pressure differences between paired holes with five differential sensors, calibrate it with a third-degree multivariate polynomial over 3-27 m/s and ±35° angles, and report mean absolute errors of 0.44 m/s for airspeed and 0.16° for angle. They also test conical versus spherical tips and four hole spacings, concluding that cone tips give better resolution and lower noise while hole spacing has no significant effect. If the claim is right, a lightweight UAV can carry one cheap, public-domain sensor instead of a pitot tube plus two vanes, and stall detection becomes feasible at low speeds.","feed_headline":"Nine-gram probe reads airflow down to 3 m/s","feed_subtitle":"A five-port differential design plus polynomial calibration estimates speed and angles across ±35 degrees.","key_machinery":"The load-bearing object is the five-channel differential pressure pattern: four pairs of peripheral holes angled at 45° plus one static-to-tip pair, each read by a differential pressure sensor, so paired holes halve the required sensor count. The key identity is the scaling factor $q=\\sqrt{\\Delta P_1^2+\\Delta P_2^2+\\Delta P_3^2+\\Delta P_4^2+\\Delta P_5^2}$; dividing each pressure difference by $q$ removes most of the airspeed dependence, and three separate third-degree polynomial regressions then map the scaled pressures to airspeed, angle of attack, and sideslip. This $q$-scaling is what lets a model trained over 3-27 m/s and ±35° generalize across the tested flight envelope, and the design comparison is what justifies the final cone-tip, 1.2 mm-spacing configuration.","core_discovery":"The paper's central claim is that a 9 g, single-component, differentially sensed multihole probe can recover airspeed, angle of attack, and sideslip across 3-27 m/s and ±35° using only a calibrated multivariate polynomial, with a mean absolute error of 0.44 m/s and 0.16° in wind-tunnel tests and consistent behavior during outdoor stall and yaw maneuvers. The authors establish this through a design comparison showing conical tips outperform spherical tips while hole spacing has no significant effect, a preprocessing step that scales five pressure differences by a common factor to remove airspeed dependence, and third-degree polynomial models selected by an elbow method. Outdoor flights against an autopilot-based estimate, rather than a true ground truth, show the probe tracks airspeed, angle of attack, and sideslip, and it keeps working during maneuvers where a pitot tube alone degrades.","pith_inferences":["The reported wind-tunnel accuracy may be optimistic because linearly interpolated artificial points are added before the train/test split, and artificial points near the zero-angle regime are nearly combinations of measured training points; retraining on the 17 measured configurations alone would reveal the honest holdout error.","The $q$-scaling preprocessing is probe-agnostic, so other multihole probes that use differential pressure sensors could adopt it, potentially making low-speed polynomial calibration a standard workflow rather than a bespoke solution.","The 50 Hz measurement rate and the polynomial inverse map could be used online to estimate turbulence intensity, but the paper only demonstrates quasi-steady calibration, so that extension remains untested.","If the sensor truly operates at 3 m/s, it may also serve as a calibration or airflow tool for very slow indoor drones, a regime the paper does not explore."],"forward_implications":["Small fixed-wing UAVs can replace a pitot tube plus two angular vanes with one 9 g probe and still obtain three airflow quantities, reducing weight and mounting complexity.","Stall detection becomes practical at low flight speeds because the angle-of-attack estimate is accurate enough to indicate approach to stall before lift is lost.","The design result tells builders that cone tips are worth using, while hole spacing can be chosen by manufacturing convenience rather than accuracy.","Because calibration is only a third-degree polynomial, the sensor can run on modest autopilot hardware without requiring a neural network.","The public release of design and calibration files allows other groups to reproduce the sensor and extend it to higher angles or other airframes."],"supporting_citations":[{"why":"shows the differential pressure-pair scheme that cuts the required sensor count in half.","marker":"[18]"},{"why":"motivates the 45° angled front holes used for angular sensitivity.","marker":"[16]"},{"why":"supplies the multivariate polynomial regression method used for calibration.","marker":"[21]"},{"why":"supplies the elbow method used to choose polynomial degree three.","marker":"[22]"},{"why":"provides an accuracy baseline for angle and speed that the probe is compared against.","marker":"[12]"},{"why":"provides an open-access five-hole probe baseline for comparison.","marker":"[13]"},{"why":"provides a miniature multihole probe baseline calibrated by neural network.","marker":"[14]"},{"why":"supports treating the central static-to-tip pair as pitot-like for axial speed estimation.","marker":"[19]"},{"why":"releases the design and calibration data underpinning the public-domain claim.","marker":"[20]"}],"fun_headline_variants":["9g probe nails airspeed and angles at slow speeds","Open-source 9g airflow sensor covers 3–27 m/s","Tiny probe tracks speed and angles near stalls","One lightweight sensor for airspeed, pitch, and yaw"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The reported accuracy rests on the assumption that adding artificial calibration points created by linear interpolation before splitting the data does not make the test error look better than it truly is.","fun_headline_variants_meta":{"raw":{"variants":["9g probe nails airspeed and angles at slow speeds","Open-source 9g airflow sensor covers 3–27 m/s","Tiny probe tracks speed and angles near stalls","One lightweight sensor for airspeed, pitch, and yaw"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000766,"raw_usage":{"total_tokens":3417,"prompt_tokens":983,"completion_tokens":2434,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":599,"completion_tokens_details":{"reasoning_tokens":2365}},"tokens_in":599,"tokens_out":2434,"duration_ms":21520,"temperature":1.0,"reasoning_tokens":2365,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T23:54:42.823439+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recalibrate all three polynomials on only the 17 measured wind-tunnel configurations, without the interpolated artificial points, then evaluate on newly measured wind-tunnel points across the same 3-27 m/s and ±35° grid; if the mean absolute error remains near 0.44 m/s and 0.16°, the central claim stands, and if it rises substantially, the headline accuracy is an artifact of the interpolation.","supporting_citations":[{"cited_title":"An unmanned aerial system for measuring profiles and turbulence in the atmospheric boundary layer,","cited_arxiv_id":null,"evidence_quote":"shows the differential pressure-pair scheme that cuts the required sensor count in half."},{"cited_title":"The use of pressure fluctuations on the nose of an aircraft for measuring air motion,","cited_arxiv_id":null,"evidence_quote":"motivates the 45° angled front holes used for angular sensitivity."},{"cited_title":"Multivariate polynomial regression in data mining: method- ology, problems and solutions,","cited_arxiv_id":null,"evidence_quote":"supplies the multivariate polynomial regression method used for calibration."},{"cited_title":"The determination of cluster number at k-mean using elbow method and purity evaluation on headline news,","cited_arxiv_id":null,"evidence_quote":"supplies the elbow method used to choose polynomial degree three."},{"cited_title":"Development of an unmanned aerial vehicle for atmo- spheric turbulence measurement,","cited_arxiv_id":null,"evidence_quote":"provides an accuracy baseline for angle and speed that the probe is compared against."},{"cited_title":"The oxford probe: an open access five- hole probe for aerodynamic measurements,","cited_arxiv_id":null,"evidence_quote":"provides an open-access five-hole probe baseline for comparison."},{"cited_title":"Miniature multihole pressure probes and their neural-network-based calibration,","cited_arxiv_id":null,"evidence_quote":"provides a miniature multihole probe baseline calibrated by neural network."},{"cited_title":"Design of pitot-static tube shapes and their influence on airspeed measurement accuracy,","cited_arxiv_id":null,"evidence_quote":"supports treating the central static-to-tip pair as pitot-like for axial speed estimation."},{"cited_title":"Miniature multihole airflow sensor for lightweight aircraft over wide speed and angular range,","cited_arxiv_id":null,"evidence_quote":"releases the design and calibration data underpinning the public-domain claim."}],"review_version":1}