{"id":"86e08ead-bcec-4cfa-ac13-8b974e5ddd3d","arxiv_id":"1908.05007","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"A new command converter and a disturbance observer improve multi-rotor acceleration tracking by using measured tilt in thrust commands and estimating external force disturbances.","lead":"This paper presents a control method that lets multi-rotor drones track a commanded acceleration more accurately, even when wind or tether forces push them around. The core idea is to compute thrust from the drone's current tilt, not the desired tilt, and add a disturbance observer that estimates and cancels external forces.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The Fig. 3 comparison validates Eq. (17) with the same pure-delay model used to derive Eq. (18), and the only flight test has no Eq. (9) baseline, so the claimed large-MOI advantage is not independently supported.","rationale":"The reader's weakest assumption identifies the same load-bearing point: the analysis depends on Eq. (10) as a pure-delay model of the attitude loop, and the simulation in Fig. 3 shares that assumption. My reading confirms that the internal derivation is coherent; the issue is not an algebraic inconsistency but an unsupported premise and a missing independent comparison. The z-channel cancellation in Eq. (17) is a genuine and robust mechanism, so the concern does not require rejection. It does require conditioning the central claim on a hardware comparison against Eq. (9), ideally with increased MOI, and on a test that does not assume the pure-delay model. Since the paper already lacks this evidence and the reader's verdict is CONDITIONAL, my stress-test supports leaving the verdict unchanged.","tokens_in":14828,"tokens_out":15971,"duration_ms":174475,"concrete_test":"Run a hardware comparison on one multirotor with a known added-inertia payload that increases the attitude delay, using the same acceleration command as in Fig. 10, and switch only the converter between Eq. (9) and Eqs. (7), (8), (17), with several repeated trials. If the z-channel RMS acceleration tracking error with the proposed converter is not substantially smaller than with Eq. (9), and if the horizontal errors are not consistent with Eq. (18), the claimed large-MOI advantage of the new converter is not established.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that Eqs. (7), (8) and (17) yield the decoupled response of Eq. (18) rests on two premises: first, that the attitude loop is well described by Eq. (10) as a pure time delay with gamma_phi = gamma_theta = gamma_h; second, that tilde_z_d(t)/tilde_z_d(t-gamma_h) is approximately 1. The z-channel benefit of Eq. (17) is robust because the measured attitude cancels the tilt denominator, but the horizontal decoupling is not: any amplitude or phase distortion in the attitude loop, or a rapidly changing tilde_z_d, changes the claimed output. The only direct comparison of Eq. (17) with the standard converter Eq. (9) is the simulation in Fig. 3, and that simulation appears to use the same Eq. (10) delay model that the derivation assumes, so it cannot falsify the pure-delay premise. The flight experiment in Fig. 10 demonstrates that the proposed converter tracks the commanded acceleration on one platform, but it does not include an Eq. (9) baseline, an added-inertia condition, or repeated trials. The large-MOI advantage claimed in the abstract and Section II is therefore supported by derivation plus a model-consistent simulation, not by an independent test.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a translational force control method for multi-rotor UAVs that combines a new acceleration-to-attitude/thrust converter with a disturbance observer (DOB). The converter, defined by Eqs. (7), (8) and (17), uses measured current roll and pitch in the thrust command. Under the assumption that actual attitude equals the desired attitude delayed by a time-varying factor (Eq. (10)), the paper derives the approximate input-output relation (Eq. (18)): horizontal acceleration channels are delayed by gamma_h while the vertical channel is undelayed, supposedly eliminating the vertical-channel degradation seen with the standard converter (Eq. (9)). A DOB is then designed, with the nominal model P_n(s) built from the attitude closed-loop transfer functions and a unity thrust channel, and its Q-filters are tuned via mu-analysis. Stability analysis models uncertainties in gain, moment of inertia, and time delay. Validation is reported through simulation (Fig. 3) comparing the proposed and standard converters under varying MOI, flight tests of acceleration tracking (Fig. 10), and DOB simulations and experiments including a tether disturbance and a wind fan.","tokens_in":15085,"tokens_out":8429,"duration_ms":81434,"significance":"If the central claims are correct, the converter would be a practically useful improvement for multi-rotors with slow attitude dynamics (large moment of inertia), and the systematic mu-analysis-based DOB design would strengthen confidence in the closed-loop system. The paper contains a clear first-principles derivation of the converter, explicitly identifies the assumptions it uses, and provides reproducible-looking simulation and experimental demonstrations. The use of measured current attitude in the thrust command is a clever and simple fix for the asynchronous realization of attitude and thrust. However, the claimed advantage over the standard converter rests on a simulation that may share the same model assumptions as the derivation, and the flight experiments do not include the large-MOI condition or a baseline comparison. The DOB nominal model appears inconsistent with the pure-delay model used to derive the converter, which undermines the robustness analysis. These issues must be resolved before the main contribution can be accepted.","major_comments":[{"comment":"The nominal model P_n(s)=Λ_n(s) is not consistent with the model used to derive the converter. Equation (10) models the attitude channels as pure time delays, and Equation (18) is derived from that delay model; however, Λ_n(s) in Equation (26) is a second-order rational transfer function. The claim that Equation (28) is 'a detailed representation' of Equation (10) is therefore unsupported, and the DOB design uses a nominal model that does not match the plant model on which the converter's performance claim rests. The μ-analysis in Equation (34), with uncertainties on K_j, J_j, and δ_j, does not cover this structural mismatch.","section":"Section III-B, Eqs. (26)-(29)"},{"comment":"The only direct comparison of the proposed converter (Eq. (17)) with the standard converter (Eq. (9)) is the simulation in Fig. 3, whose plant model is not described in the manuscript. If the simulation uses the same pure-delay model as the derivation, it cannot independently validate the decoupling claim. The flight experiment in Fig. 10 demonstrates acceleration tracking on one platform but includes no Case-1 baseline, no added-inertia condition, and no repeated trials, so the large-MOI advantage claimed in the abstract is not empirically supported.","section":"Section V, Fig. 3 and Fig. 10"},{"comment":"The approximation tilde_z_d(t)/tilde_z_d(t-γ_h)≈1 is not quantified. The paper states it is valid 'except in situations where the change in target vertical acceleration is abnormally large and rapid,' but does not provide a bound on the resulting tracking error or a characterization of the region of validity. This approximation is central to the claimed decoupling in Eq. (18); without a quantitative bound, the guaranteed performance of the converter is unclear.","section":"Section II, Eq. (18)"},{"comment":"The μ-analysis is performed on a diagonal model P_j(s) with independent uncertainties in each channel. The actual plant, as described by Eq. (12) and Eq. (18), contains coupling between channels and a ratio term that is nonlinear and time-varying; these are not represented in the uncertainty set Δ_j. Consequently, the robust stability condition in Eq. (46) does not necessarily guarantee stability for the full nonlinear system, and the stated stability margins may be optimistic.","section":"Section IV, Eqs. (33)-(47)"}],"minor_comments":[{"comment":"The organizational statements in the introduction are off by one: the text says 'In Section II, we discuss the mathematical model' but the model appears in Section I, and 'Section III deals with the force control' but that material is in Section II.","section":"Introduction, Section numbering"},{"comment":"The 'Note to Practitioners' block immediately repeats the word 'Abstract' after the heading; this should be removed or reformatted.","section":"Abstract / Note to Practitioners"},{"comment":"The sentence 'From the flight results using Equation (17) in Fig. 3' refers to a simulation figure, not flight results; the wording should be corrected to avoid confusion.","section":"Section II, solution candidate 2"},{"comment":"The procedure for determining Wδ,j(s) from 'a large amount of actual experimental data' is not described; please provide the data, a reference, or a more detailed explanation of how the bound in Eq. (41) was obtained.","section":"Section IV, Eq. (41)"},{"comment":"The comparison between estimated and measured disturbance in Fig. 12 is only qualitative; adding a quantitative metric such as RMS error would strengthen the validation of the DOB estimator.","section":"Section V, Fig. 12"},{"comment":"The wind disturbance experiment reports only single position-error values (about 1 m DOB-off versus about 0.3 m DOB-on); providing time-series plots or repeated trials would clarify trial-to-trial variability and support the claimed improvement.","section":"Section V, wind experiment"}],"recommendation":"major_revision","confidential_remarks":"The converter idea is interesting and the derivation is transparent, but the DOB nominal-model inconsistency and the lack of a non-circular validation of the central advantage are serious. I recommend a major revision. The paper would benefit from a reviewer with specific DOB expertise to assess whether the nominal-model mismatch can be repaired within the manuscript's scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe short version: this paper has one genuinely useful idea, and it's the thrust command computed from measured attitude rather than desired attitude. That's the thing to remember. The rest is a fairly standard DOB wrapper with a mu-analysis stability check, competently done.\n\nThe derivation of Eq. (17) is clear. The paper shows that if the attitude loop behaves like a pure delay gamma_h, the old converter (Eq. 9) makes the z-channel respond like z_d(t) times a ratio of cosines and delays, and that degrades when gamma_h grows. Using the measured attitude cancels the tilt denominator, so the z-channel tracks z_d(t) directly. Even if the pure-delay model is not perfect, that structural benefit for the z-channel is robust. The assumption that z_d(t)/z_d(t-gamma_h) is about 1 is reasonable for normal flight, and the authors flag it. That's honest.\n\nThe soft spot is validation. The simulation in Fig. 3 is the only direct comparison of the new converter against the old one, and it uses the same pure-delay model that the derivation assumes. So it doesn't independently confirm the model. The flight experiment in Fig. 10 shows the new converter tracks acceleration on a real platform, but there's no baseline, no added-inertia condition, and no repeated trials. The DOB experiment in Fig. 12 does compare DOB-on vs DOB-off and even shows estimated disturbance against measured force sensor, which is nice, but again single runs.\n\nThe mu-analysis is standard structured singular value stuff. The uncertainty bounds (30% MOI error, 10% gain error, 0.12 s delay) are chosen without a described procedure, and the W_delta(s) envelope is fit to the curves without explaining the fit. That's a minor concern because the envelope looks plausible, and the final tau values are taken with margins anyway. No code or data are released, which limits reproducibility.\n\nBottom line: the core idea is sound and the paper is a solid contribution to multi-rotor acceleration control. The weaknesses are in validation, not the math. I'd send it to peer review, with a request for a hardware comparison against Eq. (9), repeated trials, and better documentation of the mu-analysis choices. If you work on multi-rotor force control, especially with heavy payloads, read it—I'd cite the converter idea if I needed that trick.","headline":"A genuinely useful converter fix (thrust from measured attitude) wrapped in a standard DOB package, but the key comparison is only a simulation that shares the model's own delay assumptions.","tokens_in":15636,"tokens_out":2383,"would_cite":true,"duration_ms":24200,"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":"Computing thrust from measured current roll and pitch, rather than desired attitude, keeps a multirotor's vertical acceleration accurate even when its moment of inertia is large.","keywords":["disturbance observer","translational force control","multi-rotor UAV","acceleration tracking","kinematic inversion","moment of inertia","µ-analysis"],"falsifier":"Command a vertical acceleration step, or a sinusoid with period comparable to the attitude delay, while the multirotor is tilted by a lateral acceleration command on a platform whose moment of inertia has been increased by an added mass. If the $z$-axis tracking error grows as the vertical command gets faster or as the added mass increases, the ratio assumption $\\tilde{\\ddot{z}}_d(t)/\\tilde{\\ddot{z}}_d(t-\\gamma_h)\\approx1$ is the failing premise; if it stays flat, the pure-delay model of the attitude loop is the premise to test.","tokens_in":14536,"feed_emoji":"🚁","tokens_out":10879,"duration_ms":97743,"temperature":0.7,"pith_summary":"The paper claims that a multirotor can track a commanded three-dimensional acceleration accurately even when its attitude response is slow, by converting the desired acceleration into roll, pitch, and thrust commands differently from the standard kinematic inversion. The proposed converter computes the total thrust from the measured current attitude rather than the desired attitude, compensating the vertical thrust loss at the moment it actually occurs instead of one attitude-delay later. On the model where actual attitude is the desired attitude delayed by $\\gamma_h$, this makes the vertical channel track its command with no delay, while the horizontal channels keep only the attitude delay. Adding a disturbance observer built on the same model lets the vehicle reject external forces such as wind or payload weight. If correct, the result matters for large multirotors or ones carrying heavy cargo, where moment of inertia makes standard inversion degrade in the vertical channel and then corrupt the horizontal channels.","feed_headline":"Thrust from current attitude frees vertical tracking from delay","feed_subtitle":"It keeps z-acceleration accurate on heavy multirotors; the disturbance observer rejects wind and cargo forces.","key_machinery":"The load-bearing object is the command converter formed by Equations (7), (8), and (17): desired roll and pitch come from the usual kinematic inversion, but the total thrust is $T_{t,d}=-m\\tilde{\\ddot{z}}_d(t)/(\\cos\\varphi(t)\\cos\\theta(t))$, with $\\varphi(t)$ and $\\theta(t)$ measured from the IMU rather than taken from the delayed attitude command. That replacement is what aligns the vertical component of thrust with the current tilt, so the vertical channel no longer waits for the attitude delay. The second load-bearing mechanism is the disturbance observer—a loop that estimates the equivalent input disturbance from measured acceleration and cancels it—whose nominal model $P_n=\\Lambda_n(s)$ is built from the attitude-loop transfer functions and the instantaneous thrust transfer function, with Q-filters chosen by structured singular value analysis.","core_discovery":"On the paper's own terms, the central discovery is that replacing the desired roll and pitch in the thrust command by the measured current roll and pitch turns the multirotor's translational input–output relation into $\\tilde{\\ddot{X}}(t)\\approx[\\tilde{\\ddot{x}}_d(t-\\gamma_h),\\,\\tilde{\\ddot{y}}_d(t-\\gamma_h),\\,\\tilde{\\ddot{z}}_d(t)]$ (Equation (18)): the vertical channel tracks its command without delay, while the horizontal channels carry only the common attitude delay $\\gamma_h$. With the standard inversion of Equation (9), the thrust is computed from the desired attitude, so the actual attitude delay makes the vertical thrust compensation wrong and degrades the $z$-channel; the contamination then hurts the $x,y$ channels when an outer position loop keeps correcting. The paper further claims that the nominal model $\\Lambda_n(s)$ needed for the new converter can serve as the DOB plant, and that $\\mu$-analysis gives a less conservative stability margin ($\\tau_1>0.12$, $\\tau_2>0.09$) than a lumped small-gain bound. Flight experiments are reported to show the claimed vertical-channel accuracy and reduction of wind-induced position error from about 1 m to about 0.3 m.","pith_inferences":["A direct test of the ratio assumption in Equation (18) would be to command fast vertical oscillations; if $z$-tracking degrades with frequency, that assumption, not the converter structure, is the limit.","For an asymmetric airframe the equal-delay assumption $\\gamma_\\phi=\\gamma_\\theta=\\gamma_h$ may fail; using separately measured roll and pitch in Equation (17) is still meaningful, but the clean decoupling of Equation (18) would no longer follow.","The equivalent-input-disturbance formulation suggests the same loop can absorb slowly varying payload weight as a disturbance, which would let a delivery platform keep its tuned position gains when cargo mass changes."],"forward_implications":["For a symmetric multirotor whose roll and pitch delays are equal, the vertical acceleration channel becomes independent of the attitude delay, so increasing the moment of inertia no longer degrades $z$-axis acceleration tracking.","The horizontal channels still lag by $\\gamma_h$, so lateral acceleration commands are not tracked without delay; the converter's benefit is preventing vertical error from coupling back into them through the outer position loop.","Because the disturbance observer estimates the equivalent input force, external translational disturbances such as wind, tether pulls, or added payload weight can be cancelled without changing the position controller.","The $\\mu$-analysis stability condition permits faster Q-filters than a small-gain analysis would, allowing the DOB to reject disturbances at higher frequencies than the earlier preliminary design.","In the reported tests, actual three-dimensional acceleration follows operator commands, and wind-disturbance position error drops from about 1 m to about 0.3 m when the DOB is on."],"supporting_citations":[{"why":"Supplies the standard rigid-body translational model and the kinematic relation between roll, pitch, thrust, and acceleration that the converter inverts.","marker":"[1]"},{"why":"Represents the existing three-axis conversion approach that treats all axes simultaneously without modeling the attitude/thrust dynamics difference, the baseline the paper improves.","marker":"[4]"},{"why":"Preliminary DOB structure for translational disturbance on a hexarotor; the paper extends it with the new converter and a more accurate nominal model.","marker":"[18]"},{"why":"Provides the equivalent-input-disturbance idea that lets the actual force disturbance be represented as a signal added at the input for estimation.","marker":"[20]"},{"why":"Shows how to replace the irrational time-delay uncertainty in the plant by a rational multiplicative uncertainty weight W_delta for mu-analysis.","marker":"[24]"},{"why":"Defines structured singular value and the robust-stability condition mu<1 used to choose the Q-filter time constants.","marker":"[25]"}],"fun_headline_variants":["Thrust from current attitude eliminates vertical delay","Measured attitude in thrust: delay-free z-acceleration","Current attitude thrust: precise multirotor vertical tracking","DOB and current attitude give delay-free vertical force"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The analysis assumes the actual roll and pitch are just the desired roll and pitch delayed by the same time-varying lag $\\gamma_h$, and that the desired vertical acceleration barely changes during that lag; if the attitude response is not a pure delay or the vertical command changes abruptly, the claimed separation between vertical and horizontal tracking breaks down.","fun_headline_variants_meta":{"raw":{"variants":["Thrust from current attitude eliminates vertical delay","Measured attitude in thrust: delay-free z-acceleration","Current attitude thrust: precise multirotor vertical tracking","DOB and current attitude give delay-free vertical force"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000191,"raw_usage":{"total_tokens":1374,"prompt_tokens":1010,"completion_tokens":364,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":626,"completion_tokens_details":{"reasoning_tokens":301}},"tokens_in":626,"tokens_out":364,"duration_ms":4287,"temperature":1.0,"reasoning_tokens":301,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:25:58.445507+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Command a vertical acceleration step, or a sinusoid with period comparable to the attitude delay, while the multirotor is tilted by a lateral acceleration command on a platform whose moment of inertia has been increased by an added mass. If the $z$-axis tracking error grows as the vertical command gets faster or as the added mass increases, the ratio assumption $\\tilde{\\ddot{z}}_d(t)/\\tilde{\\ddot{z}}_d(t-\\gamma_h)\\approx1$ is the failing premise; if it stays flat, the pure-delay model of the attitude loop is the premise to test.","supporting_citations":[{"cited_title":"Quadrotor dynamics and control rev 0.1,","cited_arxiv_id":null,"evidence_quote":"Supplies the standard rigid-body translational model and the kinematic relation between roll, pitch, thrust, and acceleration that the converter inverts."},{"cited_title":"Nonlinear robust adaptive tracking control of a quadrotor uav via immersion and invariance methodology,","cited_arxiv_id":null,"evidence_quote":"Represents the existing three-axis conversion approach that treats all axes simultaneously without modeling the attitude/thrust dynamics difference, the baseline the paper improves."},{"cited_title":"Robust acceleration control of a hexarotor uav with a disturbance observer,","cited_arxiv_id":null,"evidence_quote":"Preliminary DOB structure for translational disturbance on a hexarotor; the paper extends it with the new converter and a more accurate nominal model."},{"cited_title":"Im- proving disturbance-rejection performance based on an equivalent-input- disturbance approach,","cited_arxiv_id":null,"evidence_quote":"Provides the equivalent-input-disturbance idea that lets the actual force disturbance be represented as a signal added at the input for estimation."},{"cited_title":"A robust con- trol approach for hydraulic excavators using µ-synthesis,","cited_arxiv_id":null,"evidence_quote":"Shows how to replace the irrational time-delay uncertainty in the plant by a rational multiplicative uncertainty weight W_delta for mu-analysis."},{"cited_title":"Analysis of feedback systems with structured uncertainties,","cited_arxiv_id":null,"evidence_quote":"Defines structured singular value and the robust-stability condition mu<1 used to choose the Q-filter time constants."}],"review_version":1}