Pith. sign in

REVIEW 4 major objections 4 minor 61 references

Learning Aerodynamics for the Control of Flying Humanoid Robots

T0 review · 4 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A jet-powered humanoid needs aerodynamic feedback to fly in wind: learned force models from wind-tunnel-validated CFD keep iRonCub-Mk1 balanced, where the same controller without them falls.

desk verdict A genuinely new full-stack pipeline for humanoid aerodynamics, but the headline control result is only simulation self-consistency plus a narrow Reynolds-number validity band. read the letter →

arxiv 2506.00305 v2 pith:3PJ6ZPX5 submitted 2025-05-30 cs.RO cs.LG

Antonello Paolino , Gabriele Nava , Fabio Di Natale , Fabio Bergonti , Punith Reddy Vanteddu , Donato Grassi , Luca Riccobene , Alex Zanotti
show 3 more authors
Renato Tognaccini Gianluca Iaccarino Daniele Pucci
This is my paper
classification cs.ROcs.LG
keywords aerialroboticshumanoidjet-poweredrobotaerodynamicscomputationalfluiddynamicswindtunnelexperimentsdeepneuralnetworkwhole-bodycontrol
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper sets out to show that a jet-powered humanoid robot can stay balanced under non-zero wind only if its controller knows the aerodynamic forces acting on it — and that those forces can be learned well enough to control against. The authors build the whole chain around iRonCub-Mk1, a 43 kg humanoid with jet engines on its back and forearms: wind tunnel measurements on the real robot, Reynolds-averaged Navier-Stokes (RANS) CFD simulations validated against those measurements, an automated CFD pipeline yielding about 8500 samples of per-link forces across joint and attitude configurations, and two learned models — a deep neural network and a simpler axisymmetric linear-regression model — that map pose and wind direction to forces in real time. The models feed a whole-body momentum-based controller, and in flight simulations the aerodynamic-aware controller completes the full flight envelope under non-zero relative wind, while the baseline controller that neglects aerodynamics falls. The same controller also keeps the physical robot balanced on the ground under fictitious wind profiles injected as feedback. If the pipeline holds up, jet-powered humanoids could operate outdoors where wind is the rule rather than the exception.

What carries the argument

The argument rides on two learned models and the controller that consumes them. The deep neural network maps the robot state — relative wind direction plus the 19 internal joint positions, an input in $\mathbb{R}^{22}$ — to the three force components on each link, an output in $\mathbb{R}^{39}$; it uses nine fully connected layers of 1048 ReLU neurons trained on about 8500 CFD samples with an MSE loss, reaching a normalized RME of $3\times10^{-6}$ on the validation set. The axisymmetric model treats each link as an isolated bluff body whose force areas vary sinusoidally with the link's angle of attack $\alpha_l$, with $$C_{DA}(\alpha_l)=w_0+w_1\cos\alpha_l+w_2\$sin^{2}$\alpha_l+w_3\$sin^{3}$\alpha_l+w_4\$cos^{3}$\alpha_l, \qquad C_{NA}(\alpha_l)=w_5\$sin^{2}$\alpha_l\cos\alpha_l,$$ and coefficients identified by Lasso-regularized least squares from the same CFD dataset. Link wrenches are assembled through transposed Jacobians into the centroidal momentum dynamics, where the controller adds the aerodynamic term $\Lambda_a(q,f_a)\nu$ to the feedback linearization and solves a quadratic program for thrust rates and joint velocities. Underneath both models sits the data chain that makes them usable: wind tunnel validation of RANS SST $k-\omega$ within 10% average error, and a coarse-mesh automated framework whose force error stays within 15% of experiment.

What would settle it

Measure or simulate the robot's drag force area at wind speeds clearly outside the tested 12–17 m/s band — for instance 8 m/s and 25 m/s — and compare with the Re-independent model's prediction; if the force area shifts by more than the roughly 5–10% the paper treats as negligible, the dataset and both models are misspecified for real gusts. A complementary test is to retrain the DNN with wind speed added as an input and check whether validation error drops materially.

Watch

Extended reading notes

Core claim

The central claim is that aerodynamics are both a real threat to a flying humanoid and a quantity that can be modeled well enough to compensate in real time. The paper's central comparative result comes from flight simulations over a designed envelope with non-zero relative wind: the baseline momentum-based controller, which neglects aerodynamic effects, fails and the robot falls, while the aerodynamic-aware controller, which feeds a learned aerodynamic model into the same control law, achieves balance throughout all maneuvers without failures. The claim extends to the modeling layer: the DNN and the axisymmetric model both support the task, including the robustness configuration where the simulator uses the DNN while the controller uses the simpler axisymmetric model. On the physical iRonCub-Mk1 prototype, fictitious frontal and lateral wind profiles injected into the feedback loop are counteracted by postural shifts against the estimated force, indicating that the aerodynamic feedback improves rather than disrupts the existing whole-body controller.

Load-bearing premise

The load-bearing premise is Reynolds-number independence: wind speed was left out of the CFD dataset and both learned models because the force coefficients were assumed not to change with airspeed over the studied range, so if drag actually varies with speed beyond 12–17 m/s, the controller will misestimate real gusts and lose its balance margin.

Editorial extensions

If this is right

  • In flight simulations with non-zero relative wind, the baseline controller that neglects aerodynamics fails and the robot falls, whereas the aerodynamic-aware controller completes the whole flight envelope without failures.
  • The full pipeline — wind-tunnel-validated RANS CFD, automated dataset generation, and learned per-link force models — is presented as a general methodology for any shape-shifting drone under aerodynamic disturbance, not only humanoids.
  • The simpler axisymmetric model, despite being about an order of magnitude less accurate than the DNN, is sufficient for control: in the robustness simulation the controller using it completes the envelope while the simulator uses the DNN.
  • On the physical robot, fictitious frontal wind is counteracted through ankle pitch adjustments and lateral gusts through hip and ankle roll adjustments, showing the aerodynamic feedback preserves balance on real hardware.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • My inference, beyond the paper: the Reynolds-number-independence assumption is the largest unvalidated link, since both models ignore airspeed entirely; a CFD or wind-tunnel sweep outside 12–17 m/s, or a real gust during flight, would test it directly, and adding wind speed as a DNN input is the natural fix.
  • My inference: the paper cancels aerodynamic forces as disturbances, but the same learned models could be exploited rather than fought — posture shaping to use drag, or eventual lifting surfaces — a direction the discussion itself endorses for the next design.
  • My inference: the robustness margin suggests a near-term deployment could pair the interpretable axisymmetric model with momentum-observer force estimation instead of the onboard wind sensors the paper reports as still unavailable.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper describes a complete pipeline for modeling aerodynamic forces on a jet-powered humanoid robot, iRonCub-Mk1, and for using those models in whole-body control. It reports: (i) mechanical design modifications for jet-engine integration and wind-tunnel mounting; (ii) a wind-tunnel campaign at the Politecnico di Milano GVPM facility measuring global forces and surface pressures; (iii) RANS CFD simulations validated against the wind-tunnel data, with an additional LBM comparison; (iv) an automated CFD framework that generated roughly 8500 simulations across joint configurations and wind directions; (v) a Deep Neural Network and a simpler Axisymmetric Model trained on that dataset; and (vi) a momentum-based controller that compensates aerodynamic forces, evaluated in flight simulations (baseline vs. aerodynamic-aware), in a robustness test pairing the DNN simulator with the Axisymmetric controller, and in ground balancing experiments with fictitious wind. The central claim is that the aerodynamic-aware controller enables the robot to maintain balance under non-zero relative wind, while a controller neglecting aerodynamics fails.

Significance. If the central claim is taken at face value, the paper would be a significant step for flying humanoid robots and shape-shifting drones: it combines full-scale wind-tunnel measurements of a humanoid, CFD validation, learned distributed force models, and controller integration in one pipeline. The hardware contributions (jetpack, forearms, wind-tunnel interface) and the open-source code are concrete assets. However, the validation strength is limited by three connected issues: the Reynolds-number-independence assumption is validated only over a narrow speed range, the main flight simulations are model-in-the-loop self-consistency checks, and the CFD/mesh validation covers only one configuration at one airspeed. These do not remove the value of the methodological pipeline, but they mean the paper's strongest claim about balancing under real wind is not fully established by the presented evidence.

major comments (4)
  1. [§2.2, §4.4, Eq. (4)] The Reynolds-number-independence assumption is load-bearing and is supported only over a narrow band. Section 2.2 reports wind-tunnel measurements at airspeeds of 12 and 17 m/s, with force-area variation below 5% 'throughout the remaining range' but with a measurable Re effect at high positive side-slip angles. Section 4.4 nevertheless excludes wind speed from both the DNN and the Axisymmetric Model, and Eq. (4) imposes the force magnitude through the 0.5 rho |v|^2 scaling alone. The flight-envelope wind profile in Fig. 5c and the fictitious-wind profiles in Figs. 6a and 6g include speeds outside the validated 12-17 m/s band, in particular ramps through low speeds, and bluff-body drag coefficients are known to change with Reynolds number (drag crisis). Because the simulator's truth model shares the same Re-independence assumption, no test in the paper can detect the resulting force error. The authors should either collect data at additional airspeeds or include wind speed as a model input, or else explicitly restrict all claims to the validated speed range.
  2. [§2.5, Fig. 5, Fig. 6] The flight-simulation validation is partially self-consistent. In Tests 2 and 4, the controller receives feedback from the same aerodynamic model that generates the simulated forces, so success only shows that the controller can compensate the simulator using the simulator's own estimate. The robustness test, which uses the Axisymmetric Model in the controller and the DNN in the simulator, is stronger, but both models are trained on the same CFD dataset and inherit the same Re-independence and quasi-steady assumptions, so it cannot expose systematic errors in the underlying CFD-to-model pipeline. The ground experiments in Figs. 6a-6l inject a fictitious wind into the controller without any physical aerodynamic force acting on the robot; they therefore demonstrate disturbance rejection, not aerodynamic-model accuracy. The claim that the robot 'achieves balance throughout all flight maneuvers without failures' should be presented as a simulation result under model-in-the-loop conditions, or supplemented by a hardware-in-the-loop or real-flight validation.
  3. [§2.3, §4.4, Fig. 3, Supplementary Fig. 5] The quantitative CFD validation and the mesh-sensitivity analysis are both performed for the single flight30 configuration at a single airspeed (17 m/s). The average force-area error below 10% and the mesh-induced error below 15% are therefore not demonstrated for the other wind-tunnel configurations or for the non-symmetric configurations used in the training dataset. Since the automated dataset is generated with the coarser 033C mesh, the authors should report per-configuration validation errors or otherwise justify that the flight30 result transfers across the joint-configuration space.
  4. [§4.5, Assumption 1] The neglect of aerodynamic moments acting on each link is not validated. The wind-tunnel campaign measured global force and torque, but the comparisons in Fig. 3 and the mesh-sensitivity analysis address only force areas (CDA, CLA), not moments. If local aerodynamic moments are not negligible, the momentum-based controller's compensation is incomplete. The authors should quantify the moment terms from the CFD dataset or otherwise justify the assumption with data.
minor comments (4)
  1. [§4.3 vs. §4.4] There is an inconsistency in the dataset counts: Section 4.3 describes 30 different joint configurations with 27 non-symmetric configurations, while Section 4.4 states the final dataset was generated for 24 joint configurations with 21 non-symmetric configurations. Please reconcile these numbers.
  2. [§2.4] The phrase 'maximum RME of 3×10^-6' should be defined; if it denotes root-mean-square error, use RMSE with proper units and clarify whether the value is normalized.
  3. [Figs. 5 and 6] The axis labels and units for wind velocity, aerodynamic force, and joint positions are not consistently described in the captions; please specify units (e.g., m/s, N, rad) and identify the plotted components (x/y/z) in each subfigure.
  4. [§2.2, Fig. 2d] The asymmetry observed at 12 m/s between positive and negative side-slip angles is attributed to a possible calibration error, but no further quantification or correction is given; a short discussion of how this uncertainty affects the CFD validation would be helpful.

Circularity Check

2 steps flagged · score 6.0 of 10

Self-consistent validation: flight tests use the same model in simulator and controller, and ground tests inject the model's own output as fictitious wind; balance under wind is demonstrated only against the model's own estimate.

  1. self definitional [Section 2.5, Flight Simulations (Test1–Test4); Figure 5]
    "Test1 and Test2 utilized the Axisymmetric Model to simulate aerodynamic forces, while Test3 and Test4 employed the DNN model. The tests employed different flight controllers: the Baseline Controller (Test1 and Test3), which neglects aerodynamic effects, and the Aerodynamic-aware Controller (Test2 and Test4), which uses feedback from the same aerodynamic forces model used by the corresponding simulator."

    In Tests 2 and 4 the simulated aerodynamic force that acts on the robot is generated by the same model whose output is fed to the controller's compensation term. The 'wind disturbance' and the controller's estimate of it are therefore the same function of state, so the simulation cannot reveal any aerodynamic-modeling error. The success criterion ('allow the robot to achieve balance throughout all flight maneuvers') is a consistency check between a model and its own copy, not a test of whether the learned model matches real aerodynamics. The robustness test with a DNN simulator and axisymmetric controller weakens this identity, but both models share the same CFD dataset, so it only tests model mismatch, not CFD-to-experiment error.

  2. other [Section 2.5, Ground Experiments; Figure 6]
    "we conducted tests on the real robot, focusing on balancing trials with jet engines turned off, during which we simulated an artificial fictitious wind profile directly injected into the controller as feedback, but not physically acting on the robot. These tests aim to verify the robustness and reliability of the aerodynamic-aware controller on the real robot."

    Here the 'wind' produces no aerodynamic force on the physical robot; the only signal the controller receives is the output of the same axisymmetric model that provides its feedback. The observed postural corrections therefore show the controller reacting to its own estimate injected into its input, not that the estimate corresponds to a real external force. The paper explicitly labels the wind fictitious, so this is a control-loop self-consistency test rather than independent evidence that the aerodynamic model predicts physical forces. The external content in the paper is the wind-tunnel/CFD comparison (global forces, about 10% error) and held-out DNN validation on CFD data, neither of which is invalidated by this step.

full rationale

The paper's derivation chain has substantial independent content: CFD is compared against wind-tunnel force measurements (Figs. 3a-b), the DNN is validated on a held-out CFD subset, and the robustness test uses different models in simulator and controller. However, the headline flight validation (Tests 2 and 4) and the ground experiments are self-consistent by construction: the simulated aerodynamic disturbance equals the controller's feedback model, and the ground 'wind' is only a fictitious signal injected into the controller. These steps validate the control loop's ability to cancel its own estimate, not the accuracy of the learned aerodynamic models. The Re-independence assumption (wind speed not an input) is a stated modeling limitation and a correctness risk, not an equation-level circularity. The paper also acknowledges that real-flight testing is unavailable, which appropriately limits the strength of the claim. Score 6 reflects one or more validation claims that reduce by construction while the underlying CFD-wind tunnel and held-out ML results remain independent.

Assumptions & free parameters 3 free parameters · 6 assumptions · 0 invented entities

The central modeling chain rests on several engineering assumptions: the RANS turbulence closure, incompressibility and steady flow, Reynolds-number independence, and the neglect of aerodynamic moments and inter-link interference. The only quantities fitted to data are the axisymmetric model coefficients and the neural network parameters, both trained on the same CFD dataset.

free parameters (3)
  • Axisymmetric model coefficients w0-w5 = Reported in Supplementary Table 2 (not reproduced in main text)
    Coefficients in Equations (7a)-(7b) are fit by Lasso and least-squares regression to the CFD dataset, defining CDA and CN A for every link.
  • DNN weights and biases = Not listed; 9 layers of 1048 neurons
    Trained with Adam for 60000 epochs on 80% of the CFD dataset to map joint angles and wind direction to link forces.
  • QP weights w1 and w2 = Not reported
    Hand-chosen weights in the Quadratic Program (Equation 11) trading off momentum tracking error and postural task; they affect controller behavior but not the aerodynamic model itself.
assumptions (6)
  • domain assumption RANS equations with the SST k-omega turbulence model adequately predict the mean aerodynamic forces on the robot's bluff-body shape.
    Invoked in Sections 2.3 and 4.3; validated only for the flight30 configuration at 17 m/s against wind tunnel data.
  • domain assumption The airflow around the robot is incompressible.
    Invoked in Section 4.3 and Discussion; justified by low free-stream Mach number, while the jet exhaust is near Mach 1 but kept far from the robot surfaces.
  • domain assumption Flow and aerodynamic forces are steady and quasi-steady, depending on joint angles and wind direction but not on joint velocities or accelerations.
    Stated in the Discussion as neglecting unsteady effects; the DNN inputs exclude joint velocities and accelerations.
  • ad hoc to paper Aerodynamic force coefficients are independent of Reynolds number over the operating range, so wind speed is not a model input.
    Section 2.2 explicitly assumes Re-independence to justify excluding wind speed from the CFD dataset; this is a paper-specific modeling choice.
  • ad hoc to paper Aerodynamic moments on each link are negligible compared to forces.
    Assumption 1 in Section 4.5; used to convert link forces into wrenches.
  • ad hoc to paper Each robot link can be modeled as an isolated axisymmetric body with sinusoidal drag and normal force coefficients.
    Introduced in Section 4.4 for the Axisymmetric Model; the dataset is also mirrored to enforce left-right symmetry.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Learning Aerodynamics for the Control of Flying Humanoid Robots." pith.science (2026). https://pith.science/paper/3PJ6ZPX5

@misc{pith2026250600305,
  author       = {Pith},
  title        = {Pith review of: Learning Aerodynamics for the Control of Flying Humanoid Robots},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3PJ6ZPX5}},
  note         = {Machine review of arXiv:2506.00305}
}
read the original abstract

Robots with multi-modal locomotion are an active research field due to their versatility in diverse environments. In this context, additional actuation can provide humanoid robots with aerial capabilities. Flying humanoid robots face challenges in modeling and control, particularly with aerodynamic forces. This paper addresses these challenges from a technological and scientific standpoint. The technological contribution includes the mechanical design of iRonCub-Mk1, a jet-powered humanoid robot, optimized for jet engine integration, and hardware modifications for wind tunnel experiments on humanoid robots for precise aerodynamic forces and surface pressure measurements. The scientific contribution offers a comprehensive approach to model and control aerodynamic forces using classical and learning techniques. Computational Fluid Dynamics (CFD) simulations calculate aerodynamic forces, validated through wind tunnel experiments on iRonCub-Mk1. An automated CFD framework expands the aerodynamic dataset, enabling the training of a Deep Neural Network and a linear regression model. These models are integrated into a simulator for designing aerodynamic-aware controllers, validated through flight simulations and balancing experiments on the iRonCub-Mk1 physical prototype.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

61 extracted references · 47 canonical work pages

  1. [1]

    Advanced Intelligent Systems, 2300327 (2023)

    Ramirez, J.P., Hamaza, S.: Multimodal locomotion: Next generation aerial–terrestrial mobile robotics. Advanced Intelligent Systems, 2300327 (2023)

    work page 2023

  2. [2]

    Bioinspiration & Biomimetics 10(1), 016005 (2015) https://doi.org/10.1088/1748-3190/ 10/1/016005 24

    Daler, L., Mintchev, S., Stefanini, C., Floreano, D.: A bioinspired multi-modal flying and walking robot. Bioinspiration & Biomimetics 10(1), 016005 (2015) https://doi.org/10.1088/1748-3190/ 10/1/016005 24

    doi:10.1088/1748-3190/ 2015

  3. [3]

    In: 2005 IEEE/RSJ International Conference on Intelligent Robots and Systems, pp

    Boria, F.J., Bachmann, R.J., Ifju, P.G., Quinn, R.D., Vaidyanathan, R., Perry, C., Wagener, J.: A sensor platform capable of aerial and terrestrial locomotion. In: 2005 IEEE/RSJ International Conference on Intelligent Robots and Systems, pp. 3959–3964 (2005). https://doi.org/10.1109/ IROS.2005.1545597

    arXiv 2005

  4. [4]

    Bioinspiration & Biomimetics 14(5), 056009 (2019) https://doi.org/10.1088/1748-3190/ab2ab7

    Shin, W.D., Park, J., Park, H.-W.: Development and experiments of a bio-inspired robot with multi-mode in aerial and terrestrial locomotion. Bioinspiration & Biomimetics 14(5), 056009 (2019) https://doi.org/10.1088/1748-3190/ab2ab7

    work page doi:10.1088/1748-3190/ab2ab7 2019

  5. [5]

    In: 2013 IEEE International Conference on Robotics and Automation, pp

    Kalantari, A., Spenko, M.: Design and experimental validation of hytaq, a hybrid terrestrial and aerial quadrotor. In: 2013 IEEE International Conference on Robotics and Automation, pp. 4445–4450 (2013). IEEE

    work page 2013

  6. [6]

    IEEE/CAA Journal of Automatica Sinica 6(1), 327–336 (2019) https://doi.org/10.1109/JAS.2018.7511258

    Premachandra, C., Otsuka, M., Gohara, R., Ninomiya, T., Kato, K.: A study on development of a hybrid aerial/terrestrial robot system for avoiding ground obstacles by flight. IEEE/CAA Journal of Automatica Sinica 6(1), 327–336 (2019) https://doi.org/10.1109/JAS.2018.7511258

    arXiv 2019

  7. [7]

    Nature communications 14(1), 3323 (2023)

    Sihite, E., Kalantari, A., Nemovi, R., Ramezani, A., Gharib, M.: Multi-modal mobility mor- phobot (m4) with appendage repurposing for locomotion plasticity enhancement. Nature communications 14(1), 3323 (2023)

    work page 2023

  8. [8]

    In: 2017 IEEE-RAS 17th International Conference on Humanoid Robotics (Humanoids), pp

    Huang, Z., Liu, B., Wei, J., Lin, Q., Ota, J., Zhang, Y.: Jet-hr1: Two-dimensional bipedal robot step over large obstacle based on a ducted-fan propulsion system. In: 2017 IEEE-RAS 17th International Conference on Humanoid Robotics (Humanoids), pp. 406–411 (2017). IEEE

    work page 2017

Show all 61 references
  1. [9]

    Science Robotics 6(59), 8136 (2021) https://doi.org/10.1126/ scirobotics.abf8136 https://www.science.org/doi/pdf/10.1126/scirobotics.abf8136

    Kim, K., Spieler, P., Lupu, E.-S., Ramezani, A., Chung, S.-J.: A bipedal walking robot that can fly, slackline, and skateboard. Science Robotics 6(59), 8136 (2021) https://doi.org/10.1126/ scirobotics.abf8136 https://www.science.org/doi/pdf/10.1126/scirobotics.abf8136

    2021 doi

  2. [10]

    In: 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp

    L’Erario, G., Hanover, D., Romero, A., Song, Y., Nava, G., Viceconte, P.M., Pucci, D., Scara- muzza, D.: Learning to walk and fly with adversarial motion priors. In: 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 10370–10377 (2024). IEEE

    2024

  3. [11]

    Science Robotics 2(13), 1026 (2017)

    Natale, L., Bartolozzi, C., Pucci, D., Wykowska, A., Metta, G.: icub: The not-yet-finished story of building a robot child. Science Robotics 2(13), 1026 (2017)

    2017

  4. [12]

    IEEE Robotics and Automation Letters 7(1), 526–533 (2021)

    Mohamed, H.A.O., Nava, G., L’Erario, G., Traversaro, S., Bergonti, F., Fiorio, L., Vanteddu, P.R., Braghin, F., Pucci, D.: Momentum-based extended kalman filter for thrust estimation on flying multibody robots. IEEE Robotics and Automation Letters 7(1), 526–533 (2021)

    2021

  5. [13]

    In: 2022 IEEE-RAS 21st International Conference on Humanoid Robots (Humanoids), pp

    L’Erario, G., Nava, G., Romualdi, G., Bergonti, F., Razza, V., Dafarra, S., Pucci, D.: Whole-body trajectory optimization for robot multimodal locomotion. In: 2022 IEEE-RAS 21st International Conference on Humanoid Robots (Humanoids), pp. 651–658 (2022). https://doi.org/10.110...

    2022

  6. [14]

    IEEE Transactions on Robotics 38(1), 626–645 (2021)

    Ollero, A., Tognon, M., Suarez, A., Lee, D., Franchi, A.: Past, present, and future of aerial robotic manipulators. IEEE Transactions on Robotics 38(1), 626–645 (2021)

    2021

  7. [15]

    IEEE Transactions on Robotics 33(6), 1453–1466 (2017)

    Orsag, M., Korpela, C., Bogdan, S., Oh, P.: Dexterous aerial robots—mobile manipulation using unmanned aerial systems. IEEE Transactions on Robotics 33(6), 1453–1466 (2017)

    2017

  8. [16]

    Control Engineering Practice 26, 1–10 (2014) 25

    Cabecinhas, D., Cunha, R., Silvestre, C.: A nonlinear quadrotor trajectory tracking controller with disturbance rejection. Control Engineering Practice 26, 1–10 (2014) 25

    2014

  9. [17]

    In: 2014 IEEE International Conference on Robotics and Automation (ICRA), pp

    Ruggiero, F., Cacace, J., Sadeghian, H., Lippiello, V.: Impedance control of vtol uavs with a momentum-based external generalized forces estimator. In: 2014 IEEE International Conference on Robotics and Automation (ICRA), pp. 2093–2099 (2014). IEEE

    2014

  10. [18]

    IEEE Transactions on Robotics 27(1), 129–142 (2010)

    Roberts, A., Tayebi, A.: Adaptive position tracking of vtol uavs. IEEE Transactions on Robotics 27(1), 129–142 (2010)

    2010

  11. [19]

    IEEE Transactions on Control Systems Technology26(1), 248–254 (2018) https://doi.org/10.1109/TCST.2017.2650679

    Antonelli, G., Cataldi, E., Arrichiello, F., Robuffo Giordano, P., Chiaverini, S., Franchi, A.: Adaptive trajectory tracking for quadrotor mavs in presence of parameter uncertainties and external disturbances. IEEE Transactions on Control Systems Technology26(1), 248–254 (2018...

    2018

  12. [20]

    Skipsmodelltankens meddelelse nr

    Theodorsen, T.: Theory of propellers. Skipsmodelltankens meddelelse nr. 33, April 1954 (1954)

    1954

  13. [21]

    In: Proceedings of Robotics: Science and Systems, Virtual (2021)

    Bauersfeld, L., Kaufmann, E., Foehn, P., Sun, S., Scaramuzza, D.: NeuroBEM: Hybrid Aero- dynamic Quadrotor Model. In: Proceedings of Robotics: Science and Systems, Virtual (2021). https://doi.org/10.15607/RSS.2021.XVII.042

    2021 doi

  14. [22]

    Automatica 108, 108491 (2019)

    Kai, J.-M., Hamel, T., Samson, C.: A unified approach to fixed-wing aircraft path following guidance and control. Automatica 108, 108491 (2019)

    2019

  15. [23]

    arXiv preprint arXiv:2309.13948 (2023)

    Bergonti, F., Nava, G., W¨ uest, V., Paolino, A., L’Erario, G., Pucci, D., Floreano, D.: Co- design optimisation of morphing topology and control of winged drones. arXiv preprint arXiv:2309.13948 (2023)

    2023 arXiv

  16. [24]

    Journal of environmental engineering 129(3), 222–231 (2003)

    Brown, P.P., Lawler, D.F.: Sphere drag and settling velocity revisited. Journal of environmental engineering 129(3), 222–231 (2003)

    2003

  17. [25]

    Journal of fluid mechanics 54(3), 565–575 (1972)

    Achenbach, E.: Experiments on the flow past spheres at very high reynolds numbers. Journal of fluid mechanics 54(3), 565–575 (1972)

    1972

  18. [26]

    Technical report, National Advisory Committee for Aeronautics (1922)

    Wieselsberger, C.: Further information on the laws of fluid resistance. Technical report, National Advisory Committee for Aeronautics (1922)

    1922

  19. [27]

    In: ASME Pressure Vessels and Piping Conference, vol

    Kritzinger, H., Kleijn, C., Akker, H.: Drag on a confined cylinder in axial flow. In: ASME Pressure Vessels and Piping Conference, vol. 46865, pp. 125–131 (2004)

    2004

  20. [28]

    Midland Park, NJ, 16–35 (1965)

    Hoerner, S.F.: Fluid dynamic drag. Midland Park, NJ, 16–35 (1965)

    1965

  21. [29]

    Springer, Cham (2014)

    Rieutord, M.: Fluid Dynamics: an Introduction. Springer, Cham (2014)

    2014

  22. [30]

    Handbook of numerical analysis 7, 713–1018 (2000)

    Eymard, R., Gallou¨ et, T., Herbin, R.: Finite volume methods. Handbook of numerical analysis 7, 713–1018 (2000)

    2000

  23. [31]

    Journal of fluid mechanics 238, 155–185 (1992)

    Erlebacher, G., Hussaini, M.Y., Speziale, C.G., Zang, T.A.: Toward the large-eddy simulation of compressible turbulent flows. Journal of fluid mechanics 238, 155–185 (1992)

    1992

  24. [32]

    Philosophical Transactions of the Royal Society of London IV, 123–164 (1895)

    Reynolds, O.: On the dynamical theory of incompressible viscous fluids and the determination of the criterion. Philosophical Transactions of the Royal Society of London IV, 123–164 (1895)

  25. [33]

    Computers & Fluids 24(3), 227–238 (1995) https://doi

    Shih, T.-H., Liou, W.W., Shabbir, A., Yang, Z., Zhu, J.: A new k- ε eddy viscosity model for high reynolds number turbulent flows. Computers & Fluids 24(3), 227–238 (1995) https://doi. org/10.1016/0045-7930(94)00032-T

    1995 doi

  26. [34]

    26 AIAA journal 32(8), 1598–1605 (1994)

    Menter, F.R.: Two-equation eddy-viscosity turbulence models for engineering applications. 26 AIAA journal 32(8), 1598–1605 (1994)

    1994

  27. [35]

    Journal of biomechanics 43(7), 1262–1268 (2010)

    Defraeye, T., Blocken, B., Koninckx, E., Hespel, P., Carmeliet, J.: Aerodynamic study of different cyclist positions: Cfd analysis and full-scale wind-tunnel tests. Journal of biomechanics 43(7), 1262–1268 (2010)

    2010

  28. [36]

    Journal of Wind Engineering and Industrial Aerodynamics 181, 27–45 (2018)

    Blocken, B., Druenen, T., Toparlar, Y., Andrianne, T.: Aerodynamic analysis of different cyclist hill descent positions. Journal of Wind Engineering and Industrial Aerodynamics 181, 27–45 (2018)

    2018

  29. [37]

    Sports Engineering 24, 1–11 (2021)

    Blocken, B., Gillmeier, S., Malizia, F., Druenen, T.: Impact of a motorcycle on cyclist aerodynamic drag in parallel and staggered arrangements. Sports Engineering 24, 1–11 (2021)

    2021

  30. [38]

    Journal of biomechanics 43(12), 2281–2287 (2010)

    Defraeye, T., Blocken, B., Koninckx, E., Hespel, P., Carmeliet, J.: Computational fluid dynamics analysis of cyclist aerodynamics: Performance of different turbulence-modelling and boundary- layer modelling approaches. Journal of biomechanics 43(12), 2281–2287 (2010)

    2010

  31. [39]

    PhD thesis, Universit¨ at Bayreuth, Bayreuther Graduiertenschule f¨ ur Mathematik und Naturwissenschaften-BayNAT (2023)

    Lehmann, M.: Computational study of microplastic transport at the water-air interface with a memory-optimized lattice boltzmann method. PhD thesis, Universit¨ at Bayreuth, Bayreuther Graduiertenschule f¨ ur Mathematik und Naturwissenschaften-BayNAT (2023)

    2023

  32. [42]

    In: 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp

    Vanteddu, P.R., Nava, G., Bergonti, F., L’Erario, G., Paolino, A., Pucci, D.: From cad to urdf: Co-design of a jet-powered humanoid robot including cad geometry. In: 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 11325–11331 (2024). IEEE

    2024

  33. [43]

    IEEE Transactions on Robotics 38(5), 3300–3313 (2022) https://doi.org/10.1109/TRO.2022.3170281

    Bergonti, F., Nava, G., Fiorio, L., L’Erario, G., Pucci, D.: Modeling and control of morphing covers for the adaptive morphology of humanoid robots. IEEE Transactions on Robotics 38(5), 3300–3313 (2022) https://doi.org/10.1109/TRO.2022.3170281

    2022

  34. [44]

    IEEE Robotics and Automation Letters 5(2), 2070–2077 (2020)

    L’Erario, G., Fiorio, L., Nava, G., Bergonti, F., Mohamed, H.A.O., Benenati, E., Traversaro, S., Pucci, D.: Modeling, identification and control of model jet engines for jet powered robotics. IEEE Robotics and Automation Letters 5(2), 2070–2077 (2020)

    2020

  35. [45]

    arXiv preprint arXiv:1803.04317 (2018)

    Ulvestad, A.: A brief review of current lithium ion battery technology and potential solid state battery technologies. arXiv preprint arXiv:1803.04317 (2018)

    2018 arXiv

  36. [46]

    In: 2015 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp

    Abdilla, A., Richards, A., Burrow, S.: Power and endurance modelling of battery-powered rotor- craft. In: 2015 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 675–680 (2015). IEEE

    2015

  37. [47]

    https://www.chevron.com/ -/media/chevron/operations/documents/diesel-fuel-tech-review.pdf

    Chevron Products Company: Diesel Fuels Technical Review (2007). https://www.chevron.com/ -/media/chevron/operations/documents/diesel-fuel-tech-review.pdf

    2007

  38. [48]

    In: Proceedings of the AGARD 79th Fluid Dynamics Panel Symposium, 27 Moscow, Russia (1996)

    Gibertini, G., Gasparini, L., Zasso, A.: Aerodynamic design of a civil-aeronautical low speed large wind tunnel. In: Proceedings of the AGARD 79th Fluid Dynamics Panel Symposium, 27 Moscow, Russia (1996). AGARD

    1996

  39. [49]

    AIAA journal 41(9), 1733–1742 (2003)

    Constantinescu, G., Chapelet, M., Squires, K.: Turbulence modeling applied to flow over a sphere. AIAA journal 41(9), 1733–1742 (2003)

    2003

  40. [50]

    In: 20th Numerical Towing Tank Symposium, pp

    Lopes, R., E¸ ca, L., Vaz, G.: On the decay of free-stream turbulence predicted by two-equation eddy-viscosity models. In: 20th Numerical Towing Tank Symposium, pp. 133–138 (2017)

    2017

  41. [51]

    IEEE Transactions on Systems Science and Cybernetics 5(4), 322–333 (1969)

    Fukushima, K.: Visual feature extraction by a multilayered network of analog threshold elements. IEEE Transactions on Systems Science and Cybernetics 5(4), 322–333 (1969)

    1969

  42. [52]

    IEEE Signal Processing Letters27, 1485–1489 (2020)

    Qi, J., Du, J., Siniscalchi, S.M., Ma, X., Lee, C.-H.: On mean absolute error for deep neural network based vector-to-vector regression. IEEE Signal Processing Letters27, 1485–1489 (2020)

    2020

  43. [53]

    arXiv preprint arXiv:1412.6980 (2014)

    Kingma, D.P., Ba, J.: Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 (2014)

    2014 arXiv

  44. [54]

    The journal of machine learning research15(1), 1929–1958 (2014)

    Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I., Salakhutdinov, R.: Dropout: a simple way to prevent neural networks from overfitting. The journal of machine learning research15(1), 1929–1958 (2014)

    2014

  45. [55]

    In: 2020 IEEE 17th India Council International Conference (INDICON), pp

    Badola, A., Nair, V.P., Lal, R.P.: An analysis of regularization methods in deep neural networks. In: 2020 IEEE 17th India Council International Conference (INDICON), pp. 1–6 (2020). IEEE

    2020

  46. [56]

    Advances in neural information processing systems 32 (2019)

    Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen, T., Lin, Z., Gimelshein, N., Antiga, L., et al.: Pytorch: An imperative style, high-performance deep learning library. Advances in neural information processing systems 32 (2019)

    2019

  47. [57]

    In: Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (2019)

    Akiba, T., Sano, S., Yanase, T., Ohta, T., Koyama, M.: Optuna: A next-generation hyper- parameter optimization framework. In: Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (2019)

    2019

  48. [58]

    In: 2022 International Conference on Robotics and Automation (ICRA), pp

    Hui, T., Paolino, A., Nava, G., L’Erario, G., Di Natale, F., Bergonti, F., Braghin, F., Pucci, D.: Centroidal aerodynamic modeling and control of flying multibody robots. In: 2022 International Conference on Robotics and Automation (ICRA), pp. 2017–2023 (2022). https://doi.org...

    2022

  49. [59]

    In: 2022 Sixth IEEE International Conference on Robotic Computing (IRC), pp

    Guedelha, N., Pasandi, V., L’Erario, G., Traversaro, S., Pucci, D.: A flexible matlab/simulink simulator for robotic floating-base systems in contact with the ground. In: 2022 Sixth IEEE International Conference on Robotic Computing (IRC), pp. 53–57 (2022). https://doi.org/10....

    2022

  50. [60]

    Frontiers in Robotics and AI2, 6 (2015)

    Nori, F., Traversaro, S., Eljaik, J., Romano, F., Del Prete, A., Pucci, D.: icub whole-body control through force regulation on rigid non-coplanar contacts. Frontiers in Robotics and AI2, 6 (2015)

    2015

  51. [61]

    IEEE Robotics and Automation Letters 3(1), 195–202 (2017)

    Pucci, D., Traversaro, S., Nori, F.: Momentum control of an underactuated flying humanoid robot. IEEE Robotics and Automation Letters 3(1), 195–202 (2017)

    2017

  52. [62]

    In: 2018 IEEE-RAS 18th International Conference on Humanoid Robots (Humanoids), pp

    Nava, G., Fiorio, L., Traversaro, S., Pucci, D.: Position and attitude control of an underactuated flying humanoid robot. In: 2018 IEEE-RAS 18th International Conference on Humanoid Robots (Humanoids), pp. 1–9 (2018). IEEE

    2018

  53. [63]

    Rapetti, L., Tirupachuri, Y., Darvish, K., Dafarra, S., Nava, G., Latella, C., Pucci, D.: Model- based real-time motion tracking using dynamical inverse kinematics. Algorithms 13(10) (2020) 28 https://doi.org/10.3390/a13100266 29 Data Availability The data that support the fin...

    2020 doi

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.