The field of humanoid robotics has long captivated researchers and engineers, pushing the boundaries of what machines can achieve in mimicking human capabilities.

At the forefront of this innovation, the Artificial and Mechanical Intelligence (AMI) Lab at the Italian Institute of Technology (IIT) in Genoa, Italy, has made a groundbreaking stride with the development of iRonCub3, the world’s first jet-powered flying humanoid robot.

This ambitious project, an evolution of the iCub robot, integrates advanced engineering with artificial intelligence (AI) to create a machine capable of both terrestrial locomotion and aerial mobility.

By equipping iRonCub3 with jet engines, IIT aims to redefine the role of humanoid robots in complex, real-world scenarios such as disaster relief and hazardous environment exploration.

Genesis of iRonCub3: Building on the iCub legacy

The iRonCub3 project is rooted in the iCub, a humanoid robot developed by IIT since 2004 as an open-source platform for research in embodied artificial intelligence. Standing at 104 cm tall, roughly the size of a five-year-old child, the iCub was designed to study human cognition, manipulation, and locomotion. With 53 degrees of freedom, full-body tactile sensors, and advanced AI algorithms, the iCub has been a cornerstone for robotics research, with over 40 units deployed in laboratories worldwide, from Japan to the United States.

The iRonCub3 builds on the third-generation iCub (iCub3), incorporating significant modifications to enable flight. Initiated in 2016 under the leadership of Daniele Pucci, head of the AMI Lab, the project aimed to extend the iCub’s capabilities by integrating aerial mobility.

The vision was to create a multimodal robot capable of walking, manipulating objects, and flying, thereby addressing challenges in environments where traditional robots or drones are limited. This ambitious goal required overcoming substantial engineering hurdles, from managing extreme temperatures to developing sophisticated control systems.

What is embodied artificial intelligence?

Embodied artificial intelligence refers to AI systems that learn and interact with the world through a physical body, much like humans do. Unlike traditional AI, which processes data abstractly, embodied AI emphasizes the role of physical interaction with the environment in developing cognitive abilities. For example, a robot like iCub learns by manipulating objects, mimicking how a child explores their surroundings to build understanding.

Engineering marvel: The design of iRonCub3

Jet-powered propulsion

The defining feature of iRonCub3 is its four compact jet engines, two mounted on its arms and two integrated into a jetpack on its back. These turbines generate a combined thrust exceeding 1000 newtons, enabling the 70 kg robot to lift off the ground.

However, the jet engines produce exhaust temperatures reaching 800°C and near-supersonic airflow speeds, necessitating significant modifications to the iCub’s original design. To withstand these extreme conditions, engineers developed a titanium spine and heat-resistant coverings, ensuring structural integrity and safety during flight.

The integration of jet engines posed unique challenges. Early prototypes faced risks of overheating or combustion due to the high-temperature exhaust. To address this, IIT developed strict experimental protocols and a fireproof testing chamber to safely collect data on jet engine performance.

These efforts culminated in the successful flight of iRonCub3, which lifted approximately 50 cm off the ground while maintaining stability during tests conducted in June 2025.

Advanced control systems

Achieving stable flight for a humanoid robot is a complex task due to its articulated structure and variable center of mass. Unlike symmetric drones, iRonCub3’s movable limbs create dynamic aerodynamic challenges.

To address this, IIT researchers developed a sophisticated control framework using constrained Quadratic Programming and neural networks trained on both simulated and experimental data. These AI-powered systems enable real-time adjustments to the robot’s posture and thrust, ensuring stability even in turbulent conditions.

A momentum-based trajectory planning algorithm, implemented in Python using a direct multiple-shooting approach, facilitates seamless transitions between walking and flying.

This planner, validated through simulations, is set to undergo real-world testing, marking a significant advancement in multimodal robotics. Collaborative efforts with the Polytechnic of Milan for wind tunnel testing and Stanford University for machine learning applications further enhanced the robot’s control architecture.

What is constrained Quadratic Programming?

Constrained Quadratic Programming is a mathematical optimization technique used to solve problems where a quadratic function (representing, for example, energy or error) needs to be minimized while adhering to specific constraints, such as physical limits on a robot’s movements or engine thrust. In iRonCub3, it helps balance the robot’s limbs and jet engines to maintain stable flight.

Milestone achievement: The first flight

In June 2025, iRonCub3 achieved a historic milestone by lifting 50 cm off the ground in IIT’s dedicated flight-testing area, maintaining stability throughout the maneuver. This success, detailed in a paper published in Nature Communications Engineering, followed two years of rigorous development and testing.

The flight demonstrated the robot’s ability to handle high-speed airflows and extreme temperatures while executing controlled maneuvers, a feat previously unattainable in humanoid robotics.

The achievement was not without challenges. Early prototypes, built on iCub v2.5 and v3.0 platforms, encountered issues such as tendon failures and insufficient control frequencies.

The iRonCub3 iteration addressed these by removing tendons, integrating force-torque sensors into the jetpacks, and upgrading electronics and control systems to operate at higher frequencies. These enhancements significantly improved the robot’s performance, paving the way for future testing at a dedicated facility in collaboration with Genoa Airport.

Applications and impact

Disaster relief and hazardous environments

The primary motivation behind iRonCub3 is its potential to operate in scenarios where traditional robots or drones fall short. In disaster relief, such as earthquakes or wildfires, iRonCub3’s ability to fly over obstacles and manipulate objects could prove invaluable.

For instance, it could navigate debris to locate survivors, open doors, or shut off gas valves, tasks requiring both aerial mobility and human-like dexterity. A 2023 study by the International Federation of Red Cross and Red Crescent Societies highlighted that 60% of disaster response missions involve inaccessible terrains, underscoring the need for such versatile robots.

In hazardous environments, such as nuclear facilities or chemical plants, iRonCub3’s aerial capabilities allow it to inspect areas unsafe for humans. Its teleoperation feature, built on the iCub3 platform, enables remote control, reducing risk to human operators.

Case studies from the Fukushima Daiichi nuclear disaster (2011) demonstrate the limitations of ground-based robots in navigating complex environments, highlighting the potential impact of flying humanoids.

Scientific and technological advancements

Beyond practical applications, iRonCub3 advances the field of aerial humanoid robotics. The project has pioneered new approaches to aerodynamics, thermodynamics, and control systems, with findings applicable to other robotic platforms.

For example, the thrust estimation framework developed by IIT, which uses sensor data and data-driven models, could enhance the performance of other flying robots. Additionally, the project’s open-source ethos, inherited from the iCub, encourages global collaboration, with researchers in Japan, Germany, and the UK contributing to its development.

The collaboration with Stanford University has also yielded AI models that integrate Computational Fluid Dynamics (CFD) simulations with real-time data, improving flight stability. These advancements could inform the development of future technologies, such as flying exoskeletons for human use, as envisioned by Daniele Pucci.

Future prospects and challenges

The successful flight of iRonCub3 is just the beginning. IIT plans to conduct further tests at Genoa Airport’s dedicated facility, focusing on longer flights and more complex maneuvers. However, challenges remain, including optimizing energy efficiency during flight-to-walking transitions and mitigating landing impacts.

The high energy demands of jet engines, which consume significant fuel, pose a barrier to prolonged operations, with current prototypes limited to short flights.

Moreover, the integration of AI-driven control systems with high-bandwidth actuators requires ongoing refinement to handle dynamic environments. The IIT team is exploring sensor fusion algorithms in C++ to improve localization and thrust estimation, leveraging onboard sensors like IMUs and RealSense cameras. These efforts aim to enhance iRonCub3’s autonomy, reducing reliance on teleoperation.

Human-like functionality

The iRonCub3 represents a paradigm shift in humanoid robotics, merging aerial mobility with human-like functionality. By building on the iCub platform and overcoming significant engineering challenges, IIT has set a new standard for multimodal robots.

The project’s implications extend beyond robotics, offering solutions for disaster response, hazardous inspections, and scientific research. As testing progresses and new applications emerge, iRonCub3 stands as a testament to the power of interdisciplinary innovation, bringing the vision of a flying humanoid closer to reality.

Source: icub.iit.it