pith. sign in

arxiv: 1907.05720 · v1 · pith:4J7WMMCQnew · submitted 2019-07-11 · 📡 eess.SP · cs.LG· cs.RO

Wind Estimation Using Quadcopter Motion: A Machine Learning Approach

Sam Allison , He Bai , Balaji Jayaraman This is my paper

Pith reviewed 2026-05-24 23:20 UTC · model grok-4.3

classification 📡 eess.SP cs.LGcs.RO
keywords wind estimationquadcopterLSTM neural networkmachine learningturbulencesUASDryden modellarge eddy simulation
0
0 comments X

The pith

LSTM neural network estimates turbulent wind from quadcopter motion with lower errors than wind triangle method.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a machine learning method to estimate wind velocity using only quadcopter state data like roll, pitch, and position. An LSTM network is trained on simulated flights in turbulent winds generated by Dryden and LES models. When tested, it produces estimates with smaller mean and variance errors than the traditional wind triangle approach that approximates airspeed from tilt angle. This matters for unmanned aerial systems that need wind information for navigation or sensing without carrying extra sensors. The approach shows how neural networks can learn the relationship between vehicle motion and environmental forces in simulation.

Core claim

By training a long short-term memory neural network on roll and pitch angles along with position inputs from a simulated quadcopter, the method predicts forcing wind velocities more accurately than the wind triangle approach in both Dryden gust and large eddy simulation wind fields.

What carries the argument

The LSTM neural network trained to map quadcopter states to wind velocities.

Load-bearing premise

The statistical relationship between quadcopter states and wind velocity in the simulations matches real-world conditions closely enough for the trained network to generalize.

What would settle it

A side-by-side comparison of LSTM predictions versus direct wind sensor measurements during actual outdoor quadcopter flights in turbulence.

read the original abstract

In this article, we study the well known problem of wind estimation in atmospheric turbulence using small unmanned aerial systems (sUAS). We present a machine learning approach to wind velocity estimation based on quadcopter state measurements without a wind sensor. We accomplish this by training a long short-term memory (LSTM) neural network (NN) on roll and pitch angles and quadcopter position inputs with forcing wind velocities as the targets. The datasets are generated using a simulated quadcopter in turbulent wind fields. The trained neural network is deployed to estimate the turbulent winds as generated by the Dryden gust model as well as a realistic large eddy simulation (LES) of a near-neutral atmospheric boundary layer (ABL) over flat terrain. The resulting NN predictions are compared to a wind triangle approach that uses tilt angle as an approximation of airspeed. Results from this study indicate that the LSTM-NN based approach predicts lower errors in both the mean and variance of the local wind field as compared to the wind triangle approach. The work reported in this article demonstrates the potential of machine learning for sensor-less wind estimation and has strong implications to large-scale low-altitude atmospheric sensing using sUAS for environmental and autonomous navigation applications.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper claims that an LSTM neural network trained on simulated quadcopter roll, pitch, and position data can estimate wind velocities in Dryden and LES-generated turbulent fields, achieving lower mean and variance errors than a wind-triangle baseline that uses tilt angle to approximate airspeed. All results are obtained from simulation; no real-flight data or quantitative error metrics are reported in the abstract.

Significance. If the simulation-to-real gap can be closed, the approach would offer a sensor-free wind estimation technique for sUAS that could support dense low-altitude atmospheric sampling and turbulence-aware navigation. The dual use of a simple Dryden model and a more realistic LES is a constructive element for testing robustness.

major comments (3)
  1. [Abstract] Abstract: the central claim that the LSTM-NN 'predicts lower errors in both the mean and variance' is unsupported by any numerical values, error bars, or statistical tests, rendering the performance advantage impossible to assess.
  2. [§3] §3 (simulation and training): the manuscript supplies no architecture details (layers, units, sequence length), training procedure (loss, optimizer, regularization, train/validation split), or quantitative description of how the quadcopter state trajectories were generated, all of which are required to evaluate whether the reported advantage is reproducible or an artifact of the simulation setup.
  3. [§4] §4 (results): every reported comparison is performed on trajectories generated by the identical simulated dynamics used for training; no real-flight validation or cross-validation against unmodeled effects (actuator lag, sensor noise, atmospheric spectra outside Dryden/LES) is provided, so the generalization premise remains untested.
minor comments (2)
  1. [Abstract] Clarify whether the NN inputs include only roll/pitch/position or also their derivatives or other states; the abstract is ambiguous on this point.
  2. [Figures] Add error bars or confidence intervals to any plotted error statistics so that the variance-reduction claim can be visually assessed.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for the constructive feedback. We agree that the abstract requires quantitative support and that Section 3 needs expanded implementation details; these will be added in revision. The simulation-only scope of the work is a limitation we will discuss explicitly. Point-by-point responses follow.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that the LSTM-NN 'predicts lower errors in both the mean and variance' is unsupported by any numerical values, error bars, or statistical tests, rendering the performance advantage impossible to assess.

    Authors: We agree the abstract should contain numerical results. The revised abstract will report specific mean and variance error values (with standard deviations where available) for the LSTM-NN versus the wind-triangle baseline on both Dryden and LES cases. revision: yes

  2. Referee: [§3] §3 (simulation and training): the manuscript supplies no architecture details (layers, units, sequence length), training procedure (loss, optimizer, regularization, train/validation split), or quantitative description of how the quadcopter state trajectories were generated, all of which are required to evaluate whether the reported advantage is reproducible or an artifact of the simulation setup.

    Authors: We acknowledge these details were omitted. Section 3 will be expanded to specify the LSTM architecture (layers, hidden units, sequence length), training settings (loss function, optimizer, regularization, train/validation split), and quantitative simulation parameters (quadcopter dynamics, trajectory generation, wind-field discretization). revision: yes

  3. Referee: [§4] §4 (results): every reported comparison is performed on trajectories generated by the identical simulated dynamics used for training; no real-flight validation or cross-validation against unmodeled effects (actuator lag, sensor noise, atmospheric spectra outside Dryden/LES) is provided, so the generalization premise remains untested.

    Authors: The study is deliberately simulation-based to enable controlled comparison under two turbulence models. We will add explicit discussion of this scope limitation and future real-flight plans, but cannot supply experimental data in the current revision. revision: partial

standing simulated objections not resolved
  • Absence of real-flight validation data, as the manuscript reports only simulation results and no experimental flights were performed.

Circularity Check

0 steps flagged

No significant circularity in the supervised ML pipeline on external simulation data

full rationale

The paper trains an LSTM-NN on roll/pitch/position inputs with wind velocities as targets, where all training and test trajectories are generated externally by a simulated quadcopter driven by Dryden and LES wind models. The wind-triangle baseline is an independent analytic approximation. No equation, parameter fit, or self-citation reduces the reported error reductions to an algebraic identity or definitional equivalence with the inputs; the performance comparison is an empirical outcome of standard supervised learning on held-out simulation data.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central performance claim rests on the domain assumption that simulation data captures the relevant mapping from quadcopter states to wind velocity; no free parameters or invented entities are introduced beyond standard neural-network weights.

axioms (1)
  • domain assumption Simulated quadcopter trajectories under Dryden and LES wind fields are statistically representative of real atmospheric turbulence for the purpose of training and evaluating the wind estimator.
    All training and testing data are generated from these two models; the claim that the NN generalizes therefore depends on this assumption.

pith-pipeline@v0.9.0 · 5748 in / 1377 out tokens · 24508 ms · 2026-05-24T23:20:41.065290+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

48 extracted references · 48 canonical work pages · 3 internal anchors

  1. [1]

    Exploring the Range of Weather Impacts on UAS Operations,

    Ranquist, E., Steiner, M., and Argrow, B., “Exploring the Range of Weather Impacts on UAS Operations,”18th Conference on Aviation, Range and Aerospace Meteorology, 2016, p. 11

    work page 2016

  2. [2]

    Hazards Identification and Analysis for Unmanned Aircraft System Operations,

    Belcastro, C. M., Newman, R. L., Evans, J., Klyde, D. H., Barr, L. C., and Ancel, E., “Hazards Identification and Analysis for Unmanned Aircraft System Operations,”17th AIAA Aviation Technology, Integration, and Operations Conference, 2017, p. 3269

    work page 2017

  3. [3]

    WEATHER-RELATED AVIATION ACCIDENT STUDY 2003–2007,

    FAA, “WEATHER-RELATED AVIATION ACCIDENT STUDY 2003–2007,” 2010

    work page 2003

  4. [4]

    Susceptibility of Quadcopter Flight to Turbulence,

    Solberg, J., “Susceptibility of Quadcopter Flight to Turbulence,” 2018

    work page 2018

  5. [5]

    Windfieldestimationforsmallunmannedaerialvehicles,

    Langelaan,J.W.,Alley,N.,andNeidhoefer,J.,“Windfieldestimationforsmallunmannedaerialvehicles,” JournalofGuidance Control and Dynamics, Vol. 34, No. 4, 2011, p. 1016

    work page 2011

  6. [6]

    Wind Characterization Using Onboard IMU of sUAS,

    Donnell, G. W., Feight, J. A., Lannan, N., and Jacob, J. D., “Wind Characterization Using Onboard IMU of sUAS,”2018 Atmospheric Flight Mechanics Conference, 2018, p. 2986

    work page 2018

  7. [7]

    Measuring wind with small unmanned aircraft systems,

    Prudden, S., Fisher, A., Marino, M., Mohamed, A., Watkins, S., and Wild, G., “Measuring wind with small unmanned aircraft systems,”Journal of Wind Engineering and Industrial Aerodynamics, Vol. 176, 2018, pp. 197–210. 30

    work page 2018

  8. [8]

    Overview of small fixed-wing unmanned aircraft for meteorological sampling,

    Elston, J., Argrow, B., Stachura, M., Weibel, D., Lawrence, D., and Pope, D., “Overview of small fixed-wing unmanned aircraft for meteorological sampling,”Journal of Atmospheric and Oceanic Technology, Vol. 32, No. 1, 2015, pp. 97–115

    work page 2015

  9. [9]

    Model-based Wind profiling in the Lower Atmosphere with Multirotor UAS,

    Gonzalez-Rocha, J., Woolsey, C. A., Sultan, C., and De Wekker, S. F., “Model-based Wind profiling in the Lower Atmosphere with Multirotor UAS,”AIAA Scitech 2019 Forum, 2019, p. 1598

    work page 2019

  10. [10]

    Sensing Wind from Quadrotor Motion,

    González-Rocha, J., Woolsey, C. A., Sultan, C., and De Wekker, S. F. J., “Sensing Wind from Quadrotor Motion,”Journal of Guidance, Control, and Dynamics, 2019, pp. 1–18. doi:10.2514/1.G003542, URLhttps://doi.org/10.2514/1.G003542

    work page doi:10.2514/1.g003542 2019

  11. [11]

    Computational geometric identification for quadrotor dynamics in wind fields,

    Bisheban, M., and Lee, T., “Computational geometric identification for quadrotor dynamics in wind fields,”2017 IEEE Conference on Control Technology and Applications (CCTA), IEEE, 2017, pp. 1153–1158

    work page 2017

  12. [12]

    AWindEstimationMethodwithanUnmannedRotorcraftforEnvironmental Monitoring Tasks,

    Wang, J.-Y., Luo, B., Zeng, M., andMeng, Q.-H., “AWindEstimationMethodwithanUnmannedRotorcraftforEnvironmental Monitoring Tasks,”Sensors, Vol. 18, No. 12, 2018, p. 4504

    work page 2018

  13. [13]

    Real-time wind estimation on a micro unmanned aerial vehicle using its inertial measurement unit,

    Neumann, P. P., and Bartholmai, M., “Real-time wind estimation on a micro unmanned aerial vehicle using its inertial measurement unit,”Sensors and Actuators A: Physical, Vol. 235, 2015, pp. 300–310

    work page 2015

  14. [14]

    Wind estimation in the lower atmosphere using multirotor aircraft,

    Palomaki, R. T., Rose, N. T., van den Bossche, M., Sherman, T. J., and De Wekker, S. F., “Wind estimation in the lower atmosphere using multirotor aircraft,”Journal of Atmospheric and Oceanic Technology, Vol. 34, No. 5, 2017, pp. 1183–1191

    work page 2017

  15. [15]

    Estimating Wind Velocity with a Neural Network using Quadcopter Trajectories,

    Allison, S., Bai, H., and Jayaraman, B., “Estimating Wind Velocity with a Neural Network using Quadcopter Trajectories,” AIAA Scitech 2019 Forum, 2019, p. 1596

    work page 2019

  16. [16]

    Multilayer feedforward networks are universal approximators,

    Hornik, K., Stinchcombe, M., and White, H., “Multilayer feedforward networks are universal approximators,”Neural networks, Vol. 2, No. 5, 1989, pp. 359–366

    work page 1989

  17. [17]

    Identification and control of dynamical systems using neural networks,

    Narendra, K. S., and Parthasarathy, K., “Identification and control of dynamical systems using neural networks,”IEEE Transactions on Neural Networks, Vol. 1, No. 1, 1990, pp. 4–27. doi:10.1109/72.80202

    work page doi:10.1109/72.80202 1990

  18. [18]

    Neural networks for nonlinear dynamic system modelling and identification,

    CHEN, S., and BILLINGS, S. A., “Neural networks for nonlinear dynamic system modelling and identification,”International JournalofControl ,Vol.56,No.2,1992,pp.319–346. doi:10.1080/00207179208934317,URL https://doi.org/10.1080/ 00207179208934317

    work page doi:10.1080/00207179208934317 1992

  19. [19]

    A procedure for training recurrent networks,

    Phan, M. C., Beale, M. H., and Hagan, M. T., “A procedure for training recurrent networks,”Neural Networks (IJCNN), The 2013 International Joint Conference on, IEEE, 2013, pp. 1–8

    work page 2013

  20. [20]

    Recurrent Neural Network Regularization

    Zaremba, W., Sutskever, I., and Vinyals, O., “Recurrent Neural Network Regularization,”CoRR, Vol. abs/1409.2329, 2014. URL http://arxiv.org/abs/1409.2329

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  21. [21]

    Goodfellow, I., Bengio, Y., and Courville, A.,Deep Learning, MIT Press, 2016.http://www.deeplearningbook.org

    work page 2016

  22. [22]

    Quadrotor dynamics and control rev 0.1,

    Beard, R., “Quadrotor dynamics and control rev 0.1,” 2008. 31

    work page 2008

  23. [23]

    Modelling and control of quadcopter,

    Luukkonen, T., “Modelling and control of quadcopter,”Independent research project in applied mathematics, Espoo, Vol. 22, 2011

    work page 2011

  24. [24]

    RotorS—A Modular Gazebo MAV Simulator Framework, pp

    Furrer, F., Burri, M., Achtelik, M., and Siegwart, R.,Robot Operating System (ROS): The Complete Reference (Volume 1), Springer International Publishing, Cham, 2016, Chaps. RotorS—A Modular Gazebo MAV Simulator Framework, pp. 595–625. doi:10.1007/978-3-319-26054-9_23, URL http://dx.doi.org/10.1007/978-3-319-26054-9_23

    work page doi:10.1007/978-3-319-26054-9_23 2016

  25. [25]

    Dynamiccontrolofautonomousquadrotorflightinanestimatedwindfield,

    Sydney, N., Smyth, B., andPaley, D.A., “Dynamiccontrolofautonomousquadrotorflightinanestimatedwindfield,” Decision and Control (CDC), 2013 IEEE 52nd Annual Conference on, IEEE, 2013, pp. 3609–3616

    work page 2013

  26. [26]

    A simple attitude control of quadrotor helicopter based on ziegler-nichols rules for tuning pd parameters,

    He, Z., and Zhao, L., “A simple attitude control of quadrotor helicopter based on ziegler-nichols rules for tuning pd parameters,” The Scientific World Journal, Vol. 2014, 2014

    work page 2014

  27. [27]

    Waypoint navigation of quad-rotor MAV,

    Qian, G. M., Pebrianti, D., Chun, Y. W., Hao, Y. H., and Bayuaji, L., “Waypoint navigation of quad-rotor MAV,”2017 7th IEEE International Conference on System Engineering and Technology (ICSET), IEEE, 2017, pp. 38–42

    work page 2017

  28. [28]

    Autonomous path tracking control design for a comercial quadcopter,

    Criado, R. M., and Rubio, F. R., “Autonomous path tracking control design for a comercial quadcopter,”IFAC-PapersOnLine, Vol. 48, No. 9, 2015, pp. 73–78

    work page 2015

  29. [29]

    Progress in the statistical theory of turbulence,

    Von Karman, T., “Progress in the statistical theory of turbulence,”Proceedings of the National Academy of Sciences, Vol. 34, No. 11, 1948, pp. 530–539

    work page 1948

  30. [30]

    M.,Gust loads on aircraft: concepts and applications, AIAA, 1988

    Hoblit, F. M.,Gust loads on aircraft: concepts and applications, AIAA, 1988

    work page 1988

  31. [31]

    MIL-HDBK-1791B

    Dryden, Flying qualities of piloted aircraft, Department of Defense, 2012. MIL-HDBK-1791B

    work page 2012

  32. [32]

    MIL-STD-1797A

    Dryden, Flying qualities of piloted aircraft, Department of Defense, 1990. MIL-STD-1797A

    work page 1990

  33. [33]

    Composite statistical method for modeling wind gusts,

    Schiess, J. R., “Composite statistical method for modeling wind gusts,”Journal of Aircraft, Vol. 23, No. 2, 1986, pp. 131–135

    work page 1986

  34. [34]

    Turbulent flows,

    Pope, S. B., “Turbulent flows,” , 2001

    work page 2001

  35. [35]

    Equations of Turbulent Motions of an Incompressible Turbulent Fluid,

    Kolomogorov, A., “Equations of Turbulent Motions of an Incompressible Turbulent Fluid,”Izu. Akad. Nauk SSSR Ser. Phys. VIno, 1942, pp. 1–2

    work page 1942

  36. [36]

    W., and McLain, T

    Beard, R. W., and McLain, T. W.,Small Unmanned Aircraft - Theory and Practice, Princeton University Press, 2012

    work page 2012

  37. [37]

    Transition in Atmospheric Turbulence Structure from Neutral to Convective Stability States,

    Jayaraman, B., and Brasseur, J., “Transition in Atmospheric Turbulence Structure from Neutral to Convective Stability States,” 32nd ASME Wind Energy Symposium, 2014, p. 0868

    work page 2014

  38. [38]

    The Surprising Transition in Atmospheric Boundary Layer Turbulence Structure from Neutral to Moderately Convective Stability States and Mechanisms Underlying Large-scale Rolls

    Jayaraman,B.,andBrasseur,J.G.,“TransitioninAtmosphericBoundaryLayerTurbulenceStructurefromNeutraltoModerately Convective Stability States and Implications to Large-scale Rolls,”arXiv preprint arXiv:1807.03336, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  39. [39]

    Improvement of Meteorological Data for Air Traffic Management Purposes,

    Forrester, D. A., and Dean, G. C., “Improvement of Meteorological Data for Air Traffic Management Purposes,”Air Traffic Control Quarterly, Vol. 2, No. 2, 1994, pp. 85–101. 32

    work page 1994

  40. [40]

    Modeling of urban wind field effects on unmanned rotorcraft flight,

    Galway, D., Etele, J., and Fusina, G., “Modeling of urban wind field effects on unmanned rotorcraft flight,”Journal of Aircraft, Vol. 48, No. 5, 2011, pp. 1613–1620

    work page 2011

  41. [41]

    Estimation of Atmospheric Boundary Layer Turbulence Structure using Modelled Small UAS Dynamics within LES,

    Jayaraman, B., Allison, S., and Bai, H., “Estimation of Atmospheric Boundary Layer Turbulence Structure using Modelled Small UAS Dynamics within LES,”AIAA Scitech 2019 Forum, 2019, p. 1600

    work page 2019

  42. [42]

    Alarge-eddy-simulationmodelforthestudyofplanetaryboundary-layerturbulence,

    Moeng,C.-H.,“Alarge-eddy-simulationmodelforthestudyofplanetaryboundary-layerturbulence,” JournaloftheAtmospheric Sciences, Vol. 41, No. 13, 1984, pp. 2052–2062

    work page 1984

  43. [43]

    Analysis of Monin–Obukhov similarity from large-eddy simulation,

    Khanna, S., and Brasseur, J. G., “Analysis of Monin–Obukhov similarity from large-eddy simulation,”Journal of Fluid Mechanics, Vol. 345, 1997, pp. 251–286

    work page 1997

  44. [44]

    Coherent motions in the turbulent boundary layer,

    Robinson, S. K., “Coherent motions in the turbulent boundary layer,”Annual Review of Fluid Mechanics, Vol. 23, No. 1, 1991, pp. 601–639

    work page 1991

  45. [45]

    Sparse convolution-based markov models for nonlinear fluid flows,

    Lu, C., Jayaraman, B., Whitman, J., and Chowdhary, G., “Sparse convolution-based markov models for nonlinear fluid flows,” arXiv preprint arXiv:1803.08222, 2018

    work page arXiv 2018

  46. [46]

    Deep multilayer convolution frameworks for data-driven learning of fluid flow dynamics,

    Puligilla, S. C., and Jayaraman, B., “Deep multilayer convolution frameworks for data-driven learning of fluid flow dynamics,” 2018 Fluid Dynamics Conference, 2018, p. 3091

    work page 2018

  47. [47]

    Adam: A Method for Stochastic Optimization

    Kingma, D. P., and Ba, J., “Adam: A method for stochastic optimization,”arXiv preprint arXiv:1412.6980, 2014

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  48. [48]

    A metric for covariance matrices,

    Förstner, W., and Moonen, B., “A metric for covariance matrices,”Geodesy-The Challenge of the 3rd Millennium, Springer, 2003, pp. 299–309. 33

    work page 2003