iCub

Lorenzo Natale, Chiara Bartolozzi, Francesco Nori, Giulio Sandini, Giorgio MettaOpenalex Unknown 2018 · DOI · arXiv · OpenAlex

Abstract

This paper provides a comprehensive retrospective on the iCub, one of the world’s most successful open-source humanoid research platforms. We detail the robot's 12-year evolution from a "crawling child" prototype to a sophisticated machine capable of whole-body dynamic control, large-scale tactile sensing, and cutting-edge event-driven perception. The platform's success is attributed to its open-source philosophy and its role as a standardized testbed for human-level learning and motor control.

1. Introduction

In the early 2000s, the robotics community faced a "zoo" problem: many humanoid platforms existed, but few were mature or affordable enough to become a standard. This lack of standardization hindered progress in artificial cognition, as researchers could not easily share code or benchmark algorithms across different hardware.

The iCub was designed to fill this gap. It provides a "complete" humanoid—including a human-like sensory system (vision, touch, audition) and sophisticated kinematics—specifically as a testbed for theories on human-level learning, perception, and motor control. The choice of a three-year-old child's proportions was a functional decision to study "crawling" as a precursor to complex locomotion and to facilitate natural human-robot interaction.

The iCub humanoid robot, modeled after the proportions of a three-year-old child.

2. Mechatronics and Sensory Systems

The iCub's engineering reflects a multi-disciplinary approach to humanoid design. The platform features a 53-degree-of-freedom (DoF) design. A key feature of the early versions was the use of cable-driven transmissions (tendons), which allowed for slim limbs by moving heavy motors closer to the torso, thereby reducing inertia. While these tendon drives allowed for a compact and "human-looking" design, they presented maintenance challenges for non-experts.

Later iterations (version 2.5) introduced Series Elastic Actuators (SEA) in the legs. These "C-shaped" springs provide physical compliance and accurate torque feedback, which are essential for stable bipedal walking and balancing. Furthermore, the iCub is equipped with a "skin"—a distributed system of over 4,400 capacitive tactile sensors. This allows for whole-body awareness, enabling the robot to sense contact across its entire surface rather than just at the end-effectors.

Overview of iCub's internal systems, including kinematics, head assembly, and tactile sensors.

3. Software and Control

The robot runs on the YARP (Yet Another Robot Platform) middleware, which uses a publish-subscribe architecture optimized for modularity and real-time performance across distributed clusters. Control is managed via a hierarchical whole-body strategy that prioritizes robot momentum to prevent falling while simultaneously managing lower-priority postural tasks. This architecture has enabled stable bipedal locomotion with 1 kHz control loops and an Ethernet backbone for high-speed data transmission.

4. Perception and Open-Source Impact

A significant innovation in the iCub project is the integration of Event-Driven (ED) sensing. By utilizing Dynamic Vision Sensors (DVS) instead of traditional frame-based cameras, the system only transmits data when a pixel detects a change in light. This approach reduces latency from approximately 60ms to just a few milliseconds, significantly lowering the computational load. To support research in perception, the team also released the iCub World dataset, a collection of images used to benchmark object recognition during real-time human-robot interaction.

The project's commitment to open-source hardware and software has fostered a community of over 30 laboratories worldwide, enabling code re-use and faster collective progress. As of the paper's publication, 35 iCub robots had been built and deployed globally.

5. Conclusion and Future Directions

The iCub has successfully demonstrated the power of open-source collaboration in robotics. Lessons from its deployment are shaping iCub 3.0, which aims to improve autonomy through onboard batteries (10 Ah) and Wi-Fi, simplify maintenance by transitioning from tendon drives to serial kinematics with harmonic drives, and increase power for more natural walking patterns. These advancements will ensure the iCub remains a premier platform for studying embodied cognition.

Summary updated

BibTeX
@inproceedings{natale2018icub,
  author = {Lorenzo Natale and Chiara Bartolozzi and Francesco Nori and Giulio Sandini and Giorgio Metta},
  title = {{iCub}},
  booktitle = {Openalex Unknown},
  year = {2018},
  doi = {10.1007/978-94-007-6046-2\_21},
}

Loading related content...

Figure