Pith. sign in

REVIEW 4 major objections 4 minor 21 references

iRonCub 3: The Jet-Powered Flying Humanoid Robot

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 The paper demonstrates the first liftoff of a jet-powered humanoid robot, using a whole-body controller that shifts weight support from ground contact to turbine thrust.

desk verdict First jet-powered humanoid liftoff is real and worth publishing, but the unmeasured crane tether needs addressing before the claim is fully clean. read the letter →

arxiv 2506.01125 v1 pith:X6NUMB4G submitted 2025-06-01 cs.RO

Davide Gorbani , Hosameldin Awadalla Omer Mohamed , Giuseppe L'Erario , Gabriele Nava , Punith Reddy Vanteddu , Shabarish Purushothaman Pillai , Antonello Paolino , Fabio Bergonti
show 6 more authors
Saverio Taliani Alessandro Croci Nicholas James Tremaroli Silvio Traversaro Bruno Vittorio Trombetta Daniele Pucci
This is my paper
classification cs.RO
keywords jet-poweredhumanoidrobotaerialmobilityverticaltakeoffmodelpredictivecontrolcentroidalmomentumthrustestimationwhole-bodyflight
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

This paper reports a controlled vertical liftoff of a jet-powered humanoid robot, which the authors present as the first such demonstration. The robot, iRonCub-Mk3, carries four small turbojet turbines on its forearms and a jetpack, and it leaves the ground for a few seconds during a tethered outdoor test. The result matters because it moves aerial humanoid mobility from simulation and theory to a physical feasibility point: a full human-shaped body can be lifted by thrust generated on its own structure. The paper also documents the mechanical design, the estimation and control software, and the safety infrastructure needed to run jet experiments around a humanoid.

What carries the argument

The load-bearing object is the takeoff-phase centroidal momentum equation $G[J]\dot{h} = A(q)T + \alpha m g e_3$, where $G[J]$ is the centroidal momentum matrix, $A(q)T$ collects the thrust forces, $m g e_3$ is gravity, and the scalar $\alpha$ interpolates between full ground support ($\alpha=0$) and full thrust compensation of weight ($\alpha=1$). This parameter lets the model predictive controller treat takeoff as a continuous weight-transfer problem rather than a discrete contact-switch event. The controller itself, an LPV MPC, couples this centroidal model with a nonlinear second-order model of jet thrust dynamics, and it outputs both turbine throttle and target joint positions at rates matched to the different actuators.

What would settle it

Instrument the crane tether or safety cables with a load cell during a repeat liftoff: if the measured tether tension decreases while the robot is ascending, or if the robot's z coordinate rises while the tether still carries a substantial fraction of its weight, the claim that the liftoff was produced by jet thrust alone would be refuted.

Watch

Extended reading notes

Core claim

The paper's central claim is that iRonCub-Mk3, a humanoid built on the iCub3 platform and equipped with four small turbojet turbines, achieved liftoff from the ground under jet thrust for the first time. The flight controller is a Linear Parameter Varying Model Predictive Controller that tracks a center-of-mass reference by commanding both joint positions and turbine throttle. During the takeoff phase, the controller uses the centroidal momentum equation modified with a parameter $\alpha$ that scales the fraction of body weight compensated by thrust; when $\alpha$ reaches one, the robot detaches from the ground. In the real experiment the robot maintained roll, pitch, and yaw within about five degrees, drifted roughly 20 cm in one horizontal direction and 60 cm in the other, was airborne for about three seconds, touched down, lifted off again for another three seconds, and then was automatically shut down when orientation error exceeded a threshold. Simulation tests, using the same software, showed the controller tracking a vertical takeoff and a square trajectory with low mean absolute error.

Load-bearing premise

The crane tether and loose safety cables are treated as carrying no significant vertical load, so the observed rise in the z coordinate is credited entirely to jet thrust.

Editorial extensions

If this is right

  • A humanoid-shaped body can be lifted off the ground by turbines mounted on its own forearms and back, making aerial humanoid mobility an experimental reality rather than a simulation-only concept.
  • The same software stack runs in simulation and on hardware, so the validated simulator can be used to develop more aggressive flight maneuvers before risking the physical platform.
  • Attitude can be held within roughly five degrees during a jet-powered liftoff, indicating that orientation stabilization is feasible even under turbine-induced vibration.
  • The observed horizontal drift and the double liftoff show that the next step is improving thrust and base-pose estimation rather than re-designing the mechanical platform from scratch.

Reading between the lines

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

  • A direct measurement of tether load share would settle whether the liftoff is fully self-powered; the paper does not instrument the tether, so this is the first check a skeptical reader would run.
  • The same $\alpha$ weight-transfer parameter could be extended to transition phases beyond takeoff, such as touchdown and landing, where the robot must hand weight back from thrust to ground contact.
  • Because the jet throttle loop runs at 10 Hz while joint control runs at 1000 Hz, the multi-rate coupling is a likely bottleneck for rejecting disturbances; testing with faster turbine actuation or explicit multi-rate control is a natural next experiment.
  • If the approach scales, whole-body MPC with centroidal momentum plus actuator-level thrust models could become a general template for other legged platforms equipped with unconventional thrusters.
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 reports the design, simulation, and first hardware experiments of iRonCub-Mk3, a humanoid robot equipped with four JetCat P250 Pro turbines. The methods include mechanical integration, a UKF-based thrust estimator, base-pose estimation, and a linear parameter-varying MPC. Simulation results show a takeoff and a square trajectory; the hardware experiment reports a roughly 3-second liftoff with horizontal drift, a touchdown, a second liftoff, and an automatic shutdown. The central claim is that this is the first liftoff of a jet-powered humanoid robot.

Significance. If the central claim is substantiated, the paper is a meaningful proof-of-feasibility milestone for aerial humanoid mobility. The manuscript is candid about drift, touchdown, and auto-shutdown, and the description of the outdoor test infrastructure is useful. The FEM analysis with 750 N loads and the use of the same software stack in simulation and hardware are strengths. However, the hardware result rests on the unverified assumption that the crane tether and safety cables carry negligible vertical load, and the simulation validation is incomplete because the error-metric sentence is unfinished. As written, the evidence supports an observed altitude increase under an unmeasured external support configuration, not an unassisted self-powered liftoff.

major comments (4)
  1. [III and Fig. 12] The central hardware claim is not supported by the reported measurements because the load share of the crane tether and safety cables is never quantified. Section III states that the robot is 'suspended from a 2-meter-high crane' and Fig. 12 shows cables anchored to concrete bollards, but no tether tension, cable slack, or load-cell data are reported. If the tether or cables carried any of the robot's weight, the z increase in Fig. 14 could occur with total jet thrust below body weight. The paper needs either to instrument the tether and report tension/slack during the trial, or to reduce the claim to an observed altitude increase under an unmeasured external support configuration.
  2. [IV-A] The simulation results section cuts off at 'The Mean Absolute Error (MAE)' with no numeric values, yet the conclusion in Section V states that the robot demonstrated 'accurate tracking with low mean absolute errors.' This is an unsupported claim as written. The authors should report the actual MAE values for the CoM position and orientation tracking in Fig. 13, and should avoid calling the simulation a validation in the absence of those numbers.
  3. [II-E.1 and IV-B] The alpha parameter in Eq. (2) is a controller-side compensation fraction, not a measured quantity. Alpha reaching one in Fig. 16 means the controller assumes the thrust forces fully compensate the weight term, not that an external force sensor measured zero tether load. Therefore Fig. 16 does not rule out tether assistance, and the text should not imply that alpha indicates detachment or unassisted liftoff.
  4. [IV-B] The hardware evidence consists of a single trial in which the robot 'stays airborne for roughly 3s, briefly touches down, and then lifts off again' before automatic shutdown. For a first-of-its-kind claim, one trial is suggestive but not sufficient to demonstrate a repeatable controlled liftoff; at minimum, a second trial or a clear statement of the selection criteria for reporting this trial should be provided.
minor comments (4)
  1. [III] The text describes the robot as both 'suspended from' the crane and 'securely tethered' to it; clarify whether the crane is a load-bearing suspension or a safety restraint, since this distinction is essential to interpreting the experiment.
  2. [Fig. 14] The caption labels light-blue regions as 'airborne,' but the paper does not define the criterion used to determine the airborne state (e.g., foot-contact sensors, height threshold, or video annotation); this should be stated explicitly.
  3. [II-C and II-F] The middleware name is spelled 'Y ARP' in two places; it should be 'YARP'.
  4. [II-E.1] Please report the robot's total mass and the estimated thrust-to-weight ratio; this would contextualize whether the four turbines are capable of self-lift in the first place.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the hardware liftoff is an independent observation, and the model-based components are supported by prior test-bench identification rather than by the claim itself.

full rationale

The paper's central claim is an experimental one: that iRonCub-Mk3 achieved a brief, tethered liftoff from the ground under jet thrust (Abstract; Section IV-B; Figs. 14-16). This claim rests on measured quantities — the estimated CoM height increase and the recorded turbine thrust profiles — and is not derived by construction from the controller or the simulation. The jet dynamics model used in the UKF and MPC is cited to the authors' earlier test-bench identification work [16], [17]; that work is itself an empirical system-identification effort, not an ansatz or a restatement of the present liftoff result. The simulation in Section IV-A uses the same software components and the same jet model, and therefore demonstrates internal consistency of the control stack rather than predicting the hardware outcome; but the paper does not dress the simulation up as the experimental result. The parameter alpha in Eq. 2 is a commanded compensation fraction within the take-off controller, not a fitted or measured quantity that is later renamed as a prediction; the liftoff is identified from the measured z trajectory, not from alpha reaching one. The unmeasured tether/crane load sharing is a genuine validity concern for the 'self-powered liftoff' interpretation, but it is an unsupported external assumption, not a circular reduction of the paper's derivation to its own inputs. No step in the paper reduces a claimed result to an equivalent input by definition, by fitted-parameter renaming, or by a load-bearing self-citation chain. Accordingly, no circularity steps are identified.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The central hardware demonstration depends mainly on an ad hoc takeoff parameter alpha and on thrust-model coefficients identified in the same group's prior papers. No new physical entities are introduced.

free parameters (2)
  • alpha (thrust-weight compensation parameter) = hand-scheduled 0 to 1 during takeoff
    In Eq. 2, alpha replaces unmeasured ground contact forces with a percentage of body weight compensated by thrust. It is a hand-set scheduling parameter, not measured or estimated, and determines when the robot is expected to leave the ground.
  • Jet engine model coefficients = not reported in this paper
    The second-order nonlinear jet thrust model used by the UKF and MPC was identified on a test bench in prior papers [16], [17]. The controller and estimator depend on these fitted coefficients, which are not restated or validated here.
assumptions (3)
  • domain assumption Centroidal momentum dynamics are a valid plant model for near-hover whole-body control.
    Eq. 1 uses the centroidal momentum equation from [19], which is standard but assumes the robot can be represented by its centroidal momentum during flight. This is reasonable for slow, near-hover maneuvers, which is what the paper targets.
  • domain assumption The jet engine thrust model identified on the test bench remains valid on the robot.
    Both the MPC (Section II-E) and the thrust UKF (Section II-C) use this second-order nonlinear model. The paper itself states in Section IV-B that the simplified jet model fails to capture turbine behavior, so this assumption is load-bearing and partially unmet.
  • domain assumption The base pose estimate from the IMU and depth camera is accurate enough for closed-loop control.
    Section II-D describes the UKF combining Xsens and Realsense T265; no ground-truth validation of the pose estimate is reported, and vibration-induced noise is acknowledged later.

how reviews work

0 comments
Cite this review

Pith. "Pith review of iRonCub 3: The Jet-Powered Flying Humanoid Robot." pith.science (2026). https://pith.science/paper/X6NUMB4G

@misc{pith2026250601125,
  author       = {Pith},
  title        = {Pith review of: iRonCub 3: The Jet-Powered Flying Humanoid Robot},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/X6NUMB4G}},
  note         = {Machine review of arXiv:2506.01125}
}
read the original abstract

This article presents iRonCub 3, a jet-powered humanoid robot, and its first flight experiments. Unlike traditional aerial vehicles, iRonCub 3 aims to achieve flight using a full-body humanoid form, which poses unique challenges in control, estimation, and system integration. We highlight the robot's current mechanical and software architecture, including its propulsion system, control framework, and experimental infrastructure. The control and estimation framework is first validated in simulation by performing a takeoff and tracking a reference trajectory. Then, we demonstrate, for the first time, a liftoff of a jet-powered humanoid robot - an initial but significant step toward aerial humanoid mobility. Also, we detail how the experimental area around a jet-powered humanoid robot should be designed in order to deal with a level of complexity that is substantially superior than indoor humanoid robot experiments.

Figures

Figures reproduced from arXiv: 2506.01125 by the authors.

Figure 1
Figure 1. iRonCub-Mk3 in the experimental area. A more ambitious direction is represented by the iRonCub project [10], [11], which seeks to unify aerial, terrestrial, and manipulation capabilities within a single humanoid platform. The goal is to make the iCub humanoid robot [12] fly with the addition of four jet turbines, two mounted on the arms and two on the jetpack. The potential applications of aerial humanoid robots ext… view at source ↗
Figure 2
Figure 2. CAD rendering of iRonCub-Mk3. in robotics deployment. In this article, we outline the current status of the iRonCub project, covering the latest design version, referred to as iRonCub-Mk3, a jet-powered humanoid robot developed as a research platform to investigate aerial humanoid locomotion. The iRonCub-Mk3 represents an advancement over the earlier model, iRonCub-Mk1, which was presented in the article [13]. This … view at source ↗
Figure 3
Figure 3. Exploded view (right) and assembled view (left) of the jetpack. The components labeled numerically and color coded in the figure: 1. eyelet screw, [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (11 more)
Figure 4
Figure 4. Figure 4: CAD rendering of the forearms of the iRonCub-Mk3 robot. In this version, the forearms of the iCub robot are removed to facilitate flight tests. On [PITH_FULL_IMAGE:figures/full_fig_p003_4.png]
Figure 5
Figure 5. Figure 5: Illustration of the FEM analysis on the jetpack (left) and forearm (right). Boundary conditions were fixed at the screw points on the robot, and load [PITH_FULL_IMAGE:figures/full_fig_p004_5.png]
Figure 6
Figure 6. Figure 6: Block diagram showing the interaction among the main software components. The [PITH_FULL_IMAGE:figures/full_fig_p004_6.png]
Figure 9
Figure 9. Figure 9: In green: the force-torque sensors mounted on the [PITH_FULL_IMAGE:figures/full_fig_p005_9.png]
Figure 8
Figure 8. Figure 8: In green: the force-torque sensors mounted on the [PITH_FULL_IMAGE:figures/full_fig_p005_8.png]
Figure 10
Figure 10. Figure 10: The iRonCub-MK3 robot in the Gazebo simulation environment, [PITH_FULL_IMAGE:figures/full_fig_p006_10.png]
Figure 12
Figure 12. Figure 12: The iRonCub Area includes the iRonCub-Mk3 attached to a two-meter crane. Two concrete bollards behind the robot act as anchor points, with [PITH_FULL_IMAGE:figures/full_fig_p007_12.png]
Figure 13
Figure 13. Figure 13: Simulation results depicting the tracking of the centre of mass position and base orientation performing a square trajectory. [PITH_FULL_IMAGE:figures/full_fig_p007_13.png]
Figure 14
Figure 14. Figure 14: Experimental results illustrating the tracking of the centre of mass position and orientation during take-off with the real robot. The light blue [PITH_FULL_IMAGE:figures/full_fig_p008_14.png]
Figure 15
Figure 15. Figure 15: Sequence of frames capturing the real robot performing a take-off. The robot maintains a stable orientation while exhibiting slight horizontal drift. [PITH_FULL_IMAGE:figures/full_fig_p008_15.png]
Figure 16
Figure 16. Figure 16: Estimated thrust intensities and in dotted yellow the parameter [PITH_FULL_IMAGE:figures/full_fig_p009_16.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

21 extracted references · 21 canonical work pages

  1. [16]

    Modeling, identification and control of model jet engines for jet powered robotics

    Giuseppe L’Erario, Luca Fiorio, Gabriele Nava, Fabio Bergonti, Hosameldin Awadalla Omer Mohamed, Emilio Benenati, Silvio Traver- saro, and Daniele Pucci. Modeling, identification and control of model jet engines for jet powered robotics. IEEE Robotics and Automation Letters, 5(2):2070–2077, 2020

    work page 2020

  2. [18]

    Unified multi-rate model predictive control for a jet-powered humanoid robot, 2025

    Davide Gorbani, Giuseppe L’Erario, Hosameldin Awadalla Omer Mohamed, and Daniele Pucci. Unified multi-rate model predictive control for a jet-powered humanoid robot, 2025

    work page 2025

  3. [1]

    Aerial manipulation: A literature review

    Fabio Ruggiero, Vincenzo Lippiello, and Anibal Ollero. Aerial manipulation: A literature review. IEEE Robotics and Automation Letters, 3(3):1957–1964, 2018

    work page 1957

  4. [2]

    A bipedal walking robot that can fly, slackline, and skateboard

    Kyunam Kim, Patrick Spieler, Elena-Sorina Lupu, Alireza Ramezani, and Soon-Jo Chung. A bipedal walking robot that can fly, slackline, and skateboard. Science Robotics , 6(59):eabf8136, 2021

    work page 2021

  5. [3]

    Jet-hr1: Two-dimensional bipedal robot step over large obstacle based on a ducted-fan propulsion system

    Zhifeng Huang, Biao Liu, Jiapeng Wei, Qingsheng Lin, Jun Ota, and Yun Zhang. 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) , pages 406–411, 2017

    work page 2017

  6. [4]

    Sanchez, D

    Guillermo Heredia, Antonio Enrique Jimenez-Cano, I. Sanchez, D. Llorente, V ´ıctor M. Vega, J. Braga, Jos ´e ´Angel Acosta, and Anibal Ollero. Control of a multirotor outdoor aerial manipulator. In 2014 IEEE/RSJ International Conference on Intelligent Robots and Systems , pages 3417–3422, Sep. 2014

    work page 2014

  7. [5]

    A novel hexapod robot design with flight capability

    Mark Pitonyak and Ferat Sahin. A novel hexapod robot design with flight capability. In 2017 12th System of Systems Engineering Conference (SoSE), pages 1–6, June 2017

    work page 2017

  8. [6]

    Design and experimental validation of hytaq, a hybrid terrestrial and aerial quadrotor

    Arash Kalantari and Matthew Spenko. Design and experimental validation of hytaq, a hybrid terrestrial and aerial quadrotor. In 2013 IEEE International Conference on Robotics and Automation , pages 4445–4450, May 2013

    work page 2013

Show all 21 references
  1. [7]

    Aerial and terrestrial locomotion control of lift assisted insect biobots

    Alper Bozkurt, Amit Lal, and Robert Gilmour. Aerial and terrestrial locomotion control of lift assisted insect biobots. In 2009 Annual International Conference of the IEEE Engineering in Medicine and Biology Society , pages 2058–2061, Sept 2009

    2009

  2. [8]

    A bioinspired multi-modal flying and walking robot

    Ludovic Daler, Stefano Mintchev, Cesare Stefanini, and Dario Floreano. A bioinspired multi-modal flying and walking robot. Bioinspiration & Biomimetics, 10(1):016005, 2015

    2015

  3. [9]

    A flying robot with adaptive morphology for multi-modal locomotion

    Ludovic Daler, Julien Lecoeur, Patrizia Bernadette H ¨ahlen, and Dario Floreano. A flying robot with adaptive morphology for multi-modal locomotion. In 2013 IEEE/RSJ Int. Conf. on Intelligent Robots and Systems, pages 1361–1366, Nov 2013

    2013

  4. [10]

    Momentum control of an underactuated flying humanoid robot

    Daniele Pucci, Silvio Traversaro, and Francesco Nori. Momentum control of an underactuated flying humanoid robot. IEEE Robotics and Automation Letters , 3(1):195–202, 2018

    2018

  5. [11]

    Position and attitude control of an underactuated flying humanoid robot

    Gabriele Nava, Luca Fiorio, Silvio Traversaro, and Daniele Pucci. Position and attitude control of an underactuated flying humanoid robot. In 2018 IEEE-RAS 18th International Conference on Humanoid Robots (Humanoids) , pages 1–9, 2018

    2018

  6. [12]

    icub3 avatar system: Enabling remote fully immersive embodiment of humanoid robots

    Stefano Dafarra, Ugo Pattacini, Giulio Romualdi, Lorenzo Rapetti, Riccardo Grieco, Kourosh Darvish, Gianluca Milani, Enrico Valli, Ines Sorrentino, Paolo Maria Viceconte, et al. icub3 avatar system: Enabling remote fully immersive embodiment of humanoid robots. Science Robotic...

    2024

  7. [13]

    Momentum-based extended kalman filter for thrust estimation on flying multibody robots

    Hosameldin Awadalla Omer Mohamed, Gabriele Nava, Giuseppe L’Erario, Silvio Traversaro, Fabio Bergonti, Luca Fiorio, Punith Reddy Vanteddu, Francesco Braghin, and Daniele Pucci. Momentum-based extended kalman filter for thrust estimation on flying multibody robots. IEEE Robotic...

    2022

  8. [14]

    A middle way for robotics middleware

    Paul Fitzpatrick, Elena Ceseracciu, Daniele E Domenichelli, Ali Paikan, Giorgio Metta, and Lorenzo Natale. A middle way for robotics middleware. Journal of Software Engineering for Robotics , 5(2):42–49, 2014

    2014

  9. [15]

    Nonlinear in-situ calibration of strain-gauge force/torque sensors for humanoid robots

    Hosameldin Awadalla Omer Mohamed, Gabriele Nava, Punith Reddy Vanteddu, Francesco Braghin, and Daniele Pucci. Nonlinear in-situ calibration of strain-gauge force/torque sensors for humanoid robots. In 2023 IEEE-RAS 22nd International Conference on Humanoid Robots (Humanoids), ...

    2023

  10. [17]

    Nonlinear model identification and observer design for thrust estimation of small-scale turbojet engines

    Affaf Junaid Ahamad Momin, Gabriele Nava, Giuseppe L’Erario, Hosameldin Awadalla Omer Mohamed, Fabio Bergonti, Punith Reddy Vanteddu, Francesco Braghin, and Daniele Pucci. Nonlinear model identification and observer design for thrust estimation of small-scale turbojet engines....

    2022

  11. [19]

    Centroidal dynamics of a humanoid robot

    David E Orin, Ambarish Goswami, and Sung-Hee Lee. Centroidal dynamics of a humanoid robot. Autonomous robots, 35:161–176, 2013

    2013

  12. [20]

    Design and use paradigms for gazebo, an open-source multi-robot simulator

    Nathan Koenig and Andrew Howard. Design and use paradigms for gazebo, an open-source multi-robot simulator. In 2004 IEEE/RSJ international conference on intelligent robots and systems (IROS)(IEEE Cat. No. 04CH37566) , volume 3, pages 2149–2154. Ieee, 2004

    2004

  13. [21]

    Yarp based plugins for gazebo simulator

    Enrico Mingo Hoffman, Silvio Traversaro, Alessio Rocchi, Mirko Ferrati, Alessandro Settimi, Francesco Romano, Lorenzo Natale, Antonio Bicchi, Francesco Nori, and Nikos G Tsagarakis. Yarp based plugins for gazebo simulator. In Modelling and Simulation for Autonomous Systems: Fi...

    2014

Pith tools

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