{"id":"61825fc8-611f-4608-b931-4e12537ce0f6","arxiv_id":"2412.02874","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A BEMT-based rotor thrust estimator calibrated with one scaling factor per rotor, combined with a feedforward PID thrust controller, improved wind robustness in outdoor quadcopter flights relative to the standard quadratic thrust map.","lead":"This paper tests a blade-element-momentum thrust estimator plus a feedforward PID thrust controller on two quadcopters outdoors, calibrating each rotor with only a single scaling constant instead of full aerodynamic identification. The authors report lower trajectory errors under wind than the standard quadratic thrust-to-speed map, at the cost of 15 to 20 percent more battery use.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Thrust-robustness claim rests on an estimator whose dynamic accuracy is unvalidated and on a thrust-RMSE metric that is circular: the controller minimizes error against its own estimate, not against measured thrust.","rationale":"The paper's central contribution is a generalized, minimally calibrated thrust estimator/controller whose payoff is claimed to be better robustness than the quadratic map under aerodynamic variation. That payoff must be established by accurate thrust tracking in flight or by statistically robust trajectory metrics. The static bench correlation (Pearson 0.997/0.995) shows only that the estimator is a scaled version of hover thrust; it does not establish that the same scaling or the reused BEMT coefficients describe forward flight and gust behavior. Because the low-level controller closes the loop on That, its output necessarily reduces estimated-thrust error, so reporting that error as the main thrust-tracking result is circular. This partially agrees with the reader's weakest assumption: the reader identified static-scaling extrapolation and missing in-flight ground truth, and also noted circularity; I add that the trajectory evidence itself is not significance-tested, so the presented data cannot distinguish a real aerodynamic benefit from run-to-run variation. A load-cell or wind-tunnel experiment comparing That to measured T under the reported airspeeds would settle whether the estimator premise holds. If the estimate is accurate and the trajectory improvements are significant, the claim could be accepted; as presented, it is unverified. Hence I recommend UNVERDICTED rather than CONDITIONAL, because the missing in-flight validation is essential to the central claim, not a minor revision.","tokens_in":11686,"tokens_out":9343,"duration_ms":94860,"concrete_test":"Mount a calibrated load cell between one motor and its mount on the 500 mm platform (or place the same motor/propeller/ESC in a wind tunnel with matched airflow) and fly or run it through the same waypoint path and wind conditions reported in Section V, logging both the BEMT estimate That and the measured thrust T. Apply the Section IV-A static scaling to That and compute RMSE(That−T) as a function of airspeed and gust. If the RMSE is large or grows with airspeed, the static calibration does not transfer to dynamic flight and the thrust-RMSE evidence in Fig. 10 is an artifact of closing the loop on an inaccurate estimate.","verdict_should_be":"UNVERDICTED","load_bearing_attack":"The central claim—better robustness under aerodynamic variation than the quadratic map—requires that the single static bench scaling (Section IV-A, Table III) keep the BEMT estimate faithful to actual per-rotor thrust throughout flight. This premise is never tested. The scaling is fit only at zero airspeed; the estimator's internal equations (11a)–(11c) reuse aerodynamic coefficients from a different rotor (Table I), and the small-advance-ratio/µ²≈0 decoupling in Section IV is invoked while flights reach 11.3–12.7 m/s with 3–5 m/s wind. A single multiplicative constant cannot correct dynamic errors in λs, λi, or κ if the baseline blade-element coefficients do not transfer to the 250 mm and 500 mm rotors. The thrust-tracking evidence in Figs. 7 and 10 is also circular: the feedback law (14) drives e_T = T_sp − That toward zero, and the reported thrust RMSE compares the setpoint with That from the same estimator. The open-loop thrust-to-speed arm is penalized for not tracking this estimator, so lower estimated-thrust RMSE is largely a consequence of the feedback loop, not proof that actual thrust is tracked accurately. No in-flight load cell or other independent thrust measurement is provided. The remaining position/velocity/acceleration comparisons lack significance tests and alternate with battery/wind conditions, so they cannot single-handedly carry the robustness claim.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes replacing the standard quadratic rotor thrust-to-speed map with a BEMT-based thrust estimator and a feedforward PID thrust controller for multirotor UAVs. The estimator reuses aerodynamic coefficients from the literature as a baseline and applies a per-rotor scaling factor obtained from static bench tests, avoiding per-platform aerodynamic identification. The method is implemented in PX4 and evaluated outdoors on two quadrotors (250 mm and 500 mm) by comparing trajectory and thrust RMSE against the quadratic model under wind. The authors claim that the proposed thrust estimation and control improves robustness under aerodynamically varying flight conditions.","tokens_in":12068,"tokens_out":4669,"duration_ms":50102,"significance":"If the claims hold, the practical contribution is substantial: a low-calibration, cross-platform thrust estimation and control scheme that runs in real time at 500 Hz, with an open-source PX4 implementation and experiments on two different rotor sizes. The paper also reports useful computational-load measurements. However, the significance is conditional on resolving serious validation issues: the thrust-tracking evidence is partly circular, the static scaling is extrapolated to flight conditions that violate the model assumptions, and the trajectory-level statistical claims lack significance testing.","major_comments":[{"comment":"The thrust RMSE comparison is circular. The control law in Eq. (14) feeds back the estimated thrust via e_T = T_sp - \\hat T, so the controller is explicitly designed to drive the estimated thrust to the setpoint. The thrust RMSE reported in Figs. 10a and 10b therefore measures how well the controller tracks its own feedback signal, not how accurately the actual rotor thrust follows the setpoint. No independent in-flight thrust measurement (e.g., load cell, accelerometer-derived total thrust, or wind-tunnel validation) is provided. This metric cannot support the central claim that real thrust is controlled more accurately.","section":"Section V, Eq. (14), Fig. 10"},{"comment":"The static bench scaling is extrapolated to flight regimes that violate the model assumptions. The estimator assumes a small advance ratio and sets mu^2 approximately to zero in Section IV to decouple the horizontal force, yet the reported maximum flight speeds are 11.3 m/s for the 250 mm platform and 12.7 m/s for the 500 mm platform, with wind speeds of 3.05 m/s and 4.8 m/s respectively. A single multiplicative constant fitted at zero airspeed cannot correct for dynamic errors in lambda_i, lambda_s, or kappa if the baseline aerodynamic coefficients from Table I, identified for a different motor and propeller, do not transfer to the two test rotors. The paper provides no in-flight thrust ground truth to validate this extrapolation, so the estimator's dynamic accuracy is unestablished.","section":"Section IV-A, Table III and Section V"},{"comment":"The claimed statistical improvement over the quadratic model is not supported by significance tests. The paper reports only means, medians, and standard deviations of RMSE distributions over 16 trials per method per platform. Moreover, for the 250 mm platform the mean and median acceleration RMSE in the x and y axes are slightly better with the thrust-to-speed map than with thrust control, which is acknowledged in the text but nonetheless contradicts the blanket statement that thrust control provides better robustness. Without p-values, confidence intervals, or effect-size statistics, the trajectory-level evidence is suggestive rather than conclusive.","section":"Section V, Figs. 8-9"},{"comment":"The claim that the thrust estimate is 'simply a scaled version' of the ground-truth thrust is not consistent with Table III. The ratio of estimated to measured thrust varies with throttle: for the 250 mm rotor it ranges from about 12.3 at 10% throttle to about 8.9 at 100% throttle, and for the 500 mm rotor from about 0.75 to 0.59. A high Pearson correlation is compatible with a non-constant affine or nonlinear relation, so the single-scaling-value interpretation is not uniquely supported. In addition, the implementation described in Section IV-A uses the maximum estimated thrust during flight for normalization, whereas the abstract and introduction state that a single scaling value from the bench test is used; the relationship between these two calibration procedures should be clarified.","section":"Section IV-A, Table III"}],"minor_comments":[{"comment":"The text refers to the standard thrust-to-speed map as 'Equation (5a)', but Eq. (5a) defines the aerodynamic power coefficient C_Pam; the quadratic thrust model is Eq. (1). Please correct the citation.","section":"Section V"},{"comment":"The secant update in line 13 is missing a division sign in the typeset equation; the formula should read lambda_s^{k+1} = lambda_s^k - f(lambda_s^k) * (lambda_s^k - lambda_s^{k-1}) / (f(lambda_s^k) - f(lambda_s^{k-1})).","section":"Algorithm 1"},{"comment":"Please state explicitly that the 'Throttle' column is in percent and specify whether the estimated thrust values are raw estimates from the baseline coefficients or already scaled values.","section":"Table III"},{"comment":"The calibration description is inconsistent: the introduction and Section IV-A mention a test-bench experiment producing a scaling value, while Section IV-A later states that the user need only perform an in-flight calibration by inputting maximum throttle and storing maximum estimated thrust. Please unify the description of the calibration procedure.","section":"Section IV-A"},{"comment":"The caption uses 'f-PID control' without defining the term; please spell out 'feedforward PID' in the caption or in the text.","section":"Fig. 7"}],"recommendation":"major_revision","confidential_remarks":"The paper addresses a relevant and practical problem, and the open-source PX4 implementation plus the two-platform experimental setup are commendable. My main concern is that the central validation is built on a circular thrust-tracking metric and on an unvalidated extrapolation of a static scaling law to high-speed flight. These issues are fixable in principle: the authors should either provide an independent in-flight thrust measurement or explicitly reframe the claim as a trajectory-level comparison, and they should add proper significance tests to the trajectory RMSE analysis. I therefore support major revision rather than rejection, provided the load-bearing validation points are addressed directly."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a genuine practical contribution—reusing Bangura–Mahony's BEMT estimator with one static bench scaling per rotor, plus a feedforward PID thrust loop in PX4, validated on two quadrotors outdoors. If the single-scaling generalization holds, it removes a real calibration barrier. The implementation is open source and the computational load numbers (25 µs per estimate at 500 Hz) are useful.\n\nThe static bench results show Pearson correlations of 0.997/0.995 between estimated and measured thrust across throttle, which supports the claim that the baseline coefficients give a linearly scaled estimate for these rotors. That is the strongest evidence in the paper. The outdoor trials are real: 16 per method per platform, alternating batteries, wind perpendicular to the path.\n\nWhere it gets soft: the thrust-RMSE comparison is circular. The controller minimizes error against the same estimator that produces the reported thrust output, so lower estimated-thrust RMSE is guaranteed by construction, not proof of actual thrust tracking. There is no in-flight load cell or independent thrust measurement. The single scaling constant is fitted at zero airspeed, then extrapolated to flights at 11–12.7 m/s with 3–5 m/s wind, while the model's small-advance-ratio and µ²≈0 assumptions are invoked. A single multiplicative constant can correct a static scaling error, but it is not obvious it corrects dynamic errors in inflow states. The trajectory-level RMSE improvements (position/velocity) are non-circular, but the paper reports no significance tests; the 250mm acceleration x/y axes actually favor the baseline, which the authors explain but does not strengthen the case. Power consumption rises 15–20%, which the paper discloses.\n\nOverall: the central claim—better robustness than the quadratic map—is plausible but not settled. The paper is a solid engineering report, not a definitive aerodynamic validation.\n\nWho it's for: someone building a multirotor controller who wants a drop-in thrust loop with minimal calibration, or someone working on rotor-level control allocation. It deserves a serious referee: the question of whether one static scaling suffices in dynamic flight is worth a careful experimental test, and the authors have shipped the code and data structure to enable it.","headline":"A practical single-scaling BEMT thrust controller with real two-platform flight tests; the thrust-tracking metric is circular, but the trajectory data and open-source implementation make it worth engaging.","tokens_in":12494,"tokens_out":2460,"would_cite":true,"duration_ms":22270,"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":"This paper claims that a BEMT-based closed-loop thrust estimator, calibrated by a single static bench-test scaling per rotor, can replace the standard quadratic thrust-to-speed map and improve tracking robustness under wind on small…","keywords":["thrust estimation","thrust control","multirotor UAV","quadrotor","blade element momentum theory","feedforward PID","wind disturbance","static bench calibration"],"falsifier":"Mount a thrust load cell on one rotor arm and fly the same outdoor waypoint path in wind; if the BEMT estimate calibrated by the static scaling diverges from the measured thrust as airspeed grows past the small-advance-ratio regime, the central robustness claim would not hold.","tokens_in":11513,"feed_emoji":"🚁","tokens_out":7186,"duration_ms":67612,"temperature":0.7,"pith_summary":"Multirotor autopilots usually convert rotor speed to thrust with a quadratic curve fitted on a test bench, a map that degrades in forward flight and wind. This paper argues the right place to fix that is at the rotor level, and proposes replacing the open-loop map with a closed-loop thrust estimator and controller built on blade element momentum theory (BEMT). The estimator reads each ESC's voltage, current, and speed, solves the rotor's inflow equations online, and needs only a single scaling constant per rotor from a simple static bench test. In 32 outdoor flights per platform across two very different quadrotors, the closed-loop thrust control produced lower and more consistent tracking errors in position, velocity, acceleration, and thrust than the quadratic map under wind. The payoff, if the result holds, is a low-calibration upgrade path for small multirotors that improves robustness to aerodynamic disturbances without per-platform aerodynamic identification.","feed_headline":"One static bench test tames wind for quadrotors","feed_subtitle":"Closed-loop BEMT thrust control beats the usual quadratic map in outdoor tracking tests.","key_machinery":"The carrying object is the power-balance equation of each rotor: the mechanical power from the motor, $P_m = K_q i_a \\omega$, minus the shaft friction and inertia power $P_r = I_r \\omega \\dot{\\omega}$, gives the aerodynamic power $P_{am}$, whose coefficient $C_{Pam} = P_{am}/\\omega^3$ feeds an iterative Newton–Secant solver. The solver recovers the stream inflow ratio $\\lambda_s$ by matching $C_{Pam}$ computed from ESC telemetry to the BEMT expression, then obtains the induced inflow $\\lambda_i$ from a quadratic relation, the thrust coefficient $C_T = c_1(c_2 - \\lambda)$, and hence thrust $T = C_T \\omega^2$. The generalization trick is that the $c$ and $d$ coefficients are taken from one published rotor dataset as a baseline, and a static bench test establishes that the estimate is a linear scaling of true thrust (correlation above 0.99), so one scalar per rotor calibrates the whole estimator. The control side is a feedforward PID: $u = K_{ff}T_{sp} + K_p e_T + K_i \\int e_T + K_d \\frac{de_T}{dt}$, with $e_T = T_{sp} - \\hat{T}$, running at 500 Hz.","core_discovery":"The paper's central claim is that replacing the standard static quadratic thrust-to-speed mapping with a BEMT-based closed-loop thrust estimator and feedforward PID controller improves robustness under aerodynamically varying flight conditions. The key move is to show that the BEMT model's seven aerodynamic coefficients do not need to be re-identified for each platform: reusing one published set of baseline coefficients and fitting a single scale factor from one static thrust experiment yields thrust estimates with Pearson correlations above 0.99 to ground truth for two different motor–propeller combinations. The estimated thrust is fed back at 500 Hz into a per-rotor PID with feedforward, which tracks the thrust setpoint instead of merely commanding a motor speed. Across 32 outdoor flights per platform, the thrust control gives lower mean, median, and standard deviation of RMSE for acceleration, velocity, position, and thrust than the quadratic map, with the only exceptions in some x/y acceleration axes for the smaller 250 mm platform. The result is stated as better robustness to wind, not as a general optimality claim.","pith_inferences":["If the single static scaling is carried into aggressive forward flight, the small-advance-ratio assumption that decouples horizontal force will eventually break; an instrumented rotor with an in-flight load cell could map where the scaling drifts.","The paper's own future direction — estimating the advance ratio and horizontal force directly from IMU acceleration — would remove the main modeling assumption while keeping the one-scalar calibration, and is a natural next experiment.","The measured power increase suggests the controller gains favor tracking accuracy; scheduling gains by flight phase or wind level could recover some endurance without losing the wind benefit.","The portability claim is strongest for rotors in the tested 5-inch and 13-inch range; extending to other sizes, blade counts, or coaxial layouts would test how far the single-scaling generalization reaches."],"forward_implications":["Standard autopilots can swap the quadratic thrust-to-speed map for this estimator using only propeller mass and radius plus one static bench test per rotor, removing the need for per-platform aerodynamic identification.","Tracking errors under lateral wind decrease for both a 250 mm and a 500 mm quadrotor, and the variance across runs also drops, meaning more consistent flight in gusts.","The estimator averages about 25.23 \\mu s per call and the controller about 5.29 \\mu s, so the 500 Hz closed loop runs in real time on a Pixhawk 6C.","The more aggressive rotor commands raise battery drain by 15.30% for the 250 mm platform and 20.60% for the 500 mm platform, so the benefit is a deliberate trade: precision in wind costs endurance.","On two platforms of very different size and thrust-to-weight ratio, the same calibration procedure applies unchanged, supporting the generalization claim."],"supporting_citations":[{"why":"Supplies the original BEMT thrust estimator formulation, the iterative inflow solver, and the published aerodynamic coefficient set used as the baseline.","marker":"[15]"},{"why":"Provides the BEMT derivation of propeller thrust and drag in forward flight that the estimator builds on.","marker":"[14]"},{"why":"Supplies the quadratic thrust map with a linear battery-depletion term used as the baseline comparison in flight tests.","marker":"[12]"},{"why":"Documents the helicopter-aerodynamics background and why the simple quadratic model fails in forward flight and under gusts.","marker":"[5]"},{"why":"Gives the full BEMT-based thrust estimation analysis summarized in the paper's aerodynamic modeling section.","marker":"[21]"},{"why":"Shows an alternative polynomial thrust mapping approach that the paper contrasts with its low-calibration generalization.","marker":"[11]"}],"fun_headline_variants":["One bench test sharpens rotor thrust control in wind","Closed-loop BEMT beats quadratic maps for quadrotor thrust","Single scale factor calibrates thrust for any multirotor","Better wind robustness from BEMT thrust feedback","Quadrotors fly steadier with BEMT-based thrust control"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The single per-rotor scaling constant fitted on a static bench is assumed to stay valid in dynamic flight, including forward speeds up to 12.7 m/s and gusts, and the derivation also assumes small advance ratio and small horizontal speed to drop the horizontal-force terms.","fun_headline_variants_meta":{"raw":{"variants":["One bench test sharpens rotor thrust control in wind","Closed-loop BEMT beats quadratic maps for quadrotor thrust","Single scale factor calibrates thrust for any multirotor","Better wind robustness from BEMT thrust feedback","Quadrotors fly steadier with BEMT-based thrust control"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000275,"raw_usage":{"total_tokens":1660,"prompt_tokens":980,"completion_tokens":680,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":596,"completion_tokens_details":{"reasoning_tokens":598}},"tokens_in":596,"tokens_out":680,"duration_ms":6772,"temperature":1.0,"reasoning_tokens":598,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T22:59:54.793539+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Mount a thrust load cell on one rotor arm and fly the same outdoor waypoint path in wind; if the BEMT estimate calibrated by the static scaling diverges from the measured thrust as airspeed grows past the small-advance-ratio regime, the central robustness claim would not hold.","supporting_citations":[{"cited_title":"Thrust control for multirotor aerial vehicles,","cited_arxiv_id":null,"evidence_quote":"Supplies the original BEMT thrust estimator formulation, the iterative inflow solver, and the published aerodynamic coefficient set used as the baseline."},{"cited_title":"Propeller thrust and drag in forward flight,","cited_arxiv_id":null,"evidence_quote":"Provides the BEMT derivation of propeller thrust and drag in forward flight that the estimator builds on."},{"cited_title":"Disturbance estimation and rejection for high-precision multirotor position control,","cited_arxiv_id":null,"evidence_quote":"Supplies the quadratic thrust map with a linear battery-depletion term used as the baseline comparison in flight tests."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Documents the helicopter-aerodynamics background and why the simple quadratic model fails in forward flight and under gusts."},{"cited_title":"Aerodynamics and control of quadrotors,","cited_arxiv_id":null,"evidence_quote":"Gives the full BEMT-based thrust estimation analysis summarized in the paper's aerodynamic modeling section."},{"cited_title":"Thrust mixing, saturation, and body-rate control for accurate aggressive quadrotor flight,","cited_arxiv_id":null,"evidence_quote":"Shows an alternative polynomial thrust mapping approach that the paper contrasts with its low-calibration generalization."}],"review_version":1}