CAD-Driven Co-Design for Flight-Ready Jet-Powered Humanoids
Punith Reddy Vanteddu, Davide Gorbani, Giuseppe L’Erario, Hosameldin Awadalla Omer Mohamed, Fabio Bergonti, Daniele Pucci · 2025
Read CAD-Driven Co-Design for Flight-Ready Jet-Powered Humanoids by Punith Reddy Vanteddu, Davide Gorbani, Giuseppe L’Erario, Hosameldin Awadalla Omer Mohamed, Fabio Bergonti, Daniele Pucci (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-14935.txt.
Introduction
Punith Reddy Vanteddu and five co-authors tune both the body and the flight controller of a jet-powered humanoid robot in a 2025 paper. The base robot is the iRonCub-Mk3, which must fly along hard paths. Body shape and control settings affect each other, so the team treats design as one joint task.
A planned sweep builds five thousand valid variants in CAD form. Limb lengths, jet mount angles and offsets, and mass layout all change across the set. Grouping the variants picks a small set of stand-ins for costly tests. Flight is scored on a smooth path tracked by a predictive controller. A multi-goal search then trades off the design and control goals at once.
Catalog record
| CREATOR | Punith Reddy Vanteddu, Davide Gorbani, Giuseppe L’Erario, Hosameldin Awadalla Omer Mohamed, Fabio Bergonti, Daniele Pucci |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 31 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-14935
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
CAD-Driven Co-Design for Flight-Ready Jet-Powered Humanoids Punith Reddy Vanteddu, Davide Gorbani, Giuseppe L’Erario, Hosameldin Awadalla Omer Mohamed, Fabio Bergonti, Daniele Pucci Abstract This paper presents a CAD-driven co-design framework for optimizing jet-powered aerial humanoid robots to execute dynamically constrained trajectories. Starting from the iRonCub-Mk3 model, a Design of Experiments (DoE) approach is used to generate 5,000 geometrically varied and mechanically feasible designs by modifying limb dimensions, jet interface geometry (e.g., angle and offset), and overall mass distribution. Each model is constructed through CAD assemblies to ensure structural validity and compatibility with simulation tools. To reduce computational cost and enable parameter sensitivity analysis, the models are clustered using K-means, with representative centroids selected for evaluation. A minimum-jerk trajectory is used to assess flight performance, providing position and velocity references for a momentum-based linearized Model Predictive Control (MPC) strategy. A multi-objective optimization is then conducted using the NSGA-II algorithm, jointly exploring the space of design centroids and MPC gain parameters. The objectives are to minimize trajectory tracking error and mechanical energy expenditure. The framework outputs a set of flight-ready humanoid configurations with validated control parameters, offering a structured method for selecting and implementing feasible aerial humanoid designs. I Introduction The design of aerial humanoid robots presents distinct challenges compared to both conventional aerial vehicles and ground-based humanoids. In addition to tracking and stability requirements during flight, these systems must respect anthropomorphic geometry and accommodate complex actuation schemes such as jet propulsion. The placement of thrust interfaces and the distribution of mass directly affect flight dynamics, control authority, and mechanical feasibility. As such, the design and control of aerial humanoids are tightly coupled problems that require coordinated exploration of both structural and control spaces [ 1 , 2 ] . The co-design of robotic systems—where both morphology and control are jointly optimized—has seen increasing interest across domains ranging from manipulators and grippers to legged and mobile robots. In many of these systems, morphology directly influences control performance, and co-optimization has been shown to improve task execution [ 3 , 4 , 5 ] . For example, in modular manipulators, task-specific morphology and base pose can be optimized simultaneously to improve manipulability and reduce joint effort [ 6 ] . In robotic hands, co-evolution of mechanical structure and grasping behavior using reinforcement learning enhances adaptability across diverse tasks [ 7 ] . In quadrupeds, introducing parallel compliance has been shown to increase motion robustness and explosiveness in tasks like pronking and hop-turns by coupling morphology with control planning [ 8 ] . Despite these advances, most co-design frameworks focus on systems with limited dynamic complexity or low-dimensional design spaces. Applications to humanoid platforms remain sparse, particularly in aerial systems where mass distribution and thrust interface placement are tightly coupled with control dynamics. While upper-body co-design frameworks have been developed to optimize actuator configurations for motion retargeting [ 9 ] , and mobile robot design has integrated perception, planning, and hardware constraints [ 10 ] , these methods are not directly applicable to jet-powered humanoids. The added complexity of thrust-vectoring, underactuation, and flight dynamics introduces new constraints on design feasibility and evaluation. Fig. 1 : Aerial humanoid with varied geometric assemblies obtained from a co-design framework. Another challenge is the computational cost of exploring high-dimensional design spaces where each configuration must satisfy structural feasibility and dynamic requirements. Frameworks that use parallel training [ 7 ] , differentiable simulation [ 11 ] , or trajectory optimization [ 12 ] address this issue in part but often rely on exploratory design loops that may not guarantee manufacturable outcomes. In contrast, recent work in legged robot compliance design [ 12 ] , morphology-driven robustness [ 8 ] , and robust multi-objective optimization [ 13 , 14 ] emphasizes solution feasibility and tolerance under uncertainty. However, these methods have not yet been extended to jet-powered humanoids, where morphology and trajectory performance must be optimized jointly under fabrication constraints. Work has also been done to integrate CAD-informed geometry into co-design pipelines for jet-powered humanoids. These approaches focus on identifying control-relevant link parameters and filtering non-viable candidates using FEM-based structural analysis [ 15 ] . Such pipelines validate the utility of CAD-informed modeling in co-design but primarily target component-level optimization and do not incorporate trajectory-level performance metrics across large-scale morphological variation. In this work, we propose a framework that scales the design space and evaluates full-body CAD-derived configurations through flight-based performance metrics. Rather than optimizing isolated links, we vary whole-body geometry across thousands of models and assess each using a predictive controller in dynamic flight scenarios. The objective is to identify not only structurally feasible robots but also those capable of executing energy-efficient, dynamically constrained aerial trajectories. We present a CAD-driven co-design framework for aerial humanoid robots that integrates parametric geometry generation, structured sampling via Design of Experiments (DoE) [ 16 ] , and trajectory-based evaluation using model predictive control. Starting from the iRonCub-Mk3 design [ 17 , 18 ] , we generate 5,000 physically feasible robot configurations by varying geometric parameters, such as link lengths and turbine placements, directly extracted from CAD. To reduce computational load, we apply K-means clustering [ 19 ] and select representative centroids. Each centroid is evaluated on a minimum-jerk trajectory using a linearized momentum-based MPC. A multi-objective optimization using NSGA-II [ 20 ] jointly searches for morphology and tunes control gain parameters to minimize trajectory tracking error and energy consumption. The remainder of this paper is organized as follows: Section II reviews background and related work. Section III details the proposed framework, including model generation, trajectory definition, control formulation, and optimization procedure. Section IV presents the results, and Section V concludes the paper with discussions and future directions. II Background Fig. 2 : CAD assemblies of the links being modified. 1: Jetpack Turbine Angle; 2: Jetpack Turbine offset distance; 3: Jetpack Turbine height; 4: Forearm length; 5: Shoulder width; 6: Hip distance; 7: Ankle link height; 8: Foot length II-A Notation • S(x) : Skew-symmetric matrix such that x\times y=S(x)y for x,y\in\mathbb{R}^{3} . • \text{SO}(3) : Special orthogonal group of 3D rotations. • \mathcal{I} : Inertial frame; \mathcal{B} : Base frame attached to the robot. • \mathcal{G}[\mathcal{I}] : Frame with origin at the CoM and orientation of \mathcal{I} . • R_{\mathcal{B}}\in\text{SO}(3) : Rotation matrix from base frame to inertial frame. • {}^{\mathcal{G}[\mathcal{I}]}h : Centroidal momentum, defined as [{}^{\mathcal{G}[\mathcal{I}]}l;\,{}^{\mathcal{G}[\mathcal{I}]}w] , where l is linear and w is angular momentum. • {}^{\mathcal{G}[\mathcal{B}]}l , {}^{\mathcal{G}[\mathcal{B}]}w : Momentum terms expressed in body frame. • \mathbf{x} : MPC state vector. • \mathbf{u} : MPC input vector, composed of joint displacement \Delta\mathbf{s} and turbine throttle u_{th} . II-B Robot Model and Flight Control The robot used in this study is based on the iRonCub-Mk3 aerial humanoid, modeled as a floating-base multibody system of n+1 rigid links and n actuated joints. Turbine thrusters are rigidly mounted on specific links and provide thrust forces during flight. The configuration vector is defined as q:=(p_{B},R_{B},s) , where p_{B}\in\mathbb{R}^{3} and R_{B}\in\text{SO}(3) define the base position and orientation, and s\in\mathbb{R}^{n} the joint angles. The velocity vector is \nu:=(\dot{p}_{B},\omega_{B},\dot{s}) . The system dynamics are formulated in terms of centroidal momentum, capturing the influence of all external forces. The total momentum is given by: {}^{\mathcal{G}[\mathcal{I}]}h=\begin{bmatrix}{}^{\mathcal{G}[\mathcal{I}]}l\\ {}^{\mathcal{G}[\mathcal{I}]}w\end{bmatrix},\quad{}^{\mathcal{G}[\mathcal{I}]}\dot{h}=\begin{bmatrix}mge_{3}+\sum_{k=1}^{n_{p}}F_{k}\\ \sum_{k=1}^{n_{p}}S(r_{k})F_{k}\end{bmatrix} where g