Researchers from the Faculty of Electrical Engineering at the Czech Technical University in Prague used the humanoid robot iCub as a model for studying one of the fundamental abilities acquired during the first months of life: understanding that our own movements can influence the world around us. The experiment brings together robotics and artificial intelligence with developmental psychology and neuroscience. Its primary aim is not to teach the robot a new practical task, but to use a robotic model to better understand the mechanisms underlying the early development of human cognition.
The team led by Associate Professor Matěj Hoffmann from the Department of Cybernetics at FEE CTU built on the classic work of American psychologist Carolyn Rovee-Collier. In an experiment known as the “mobile paradigm,” an infant lies in a crib with one of its limbs connected to a suspended mobile. When the child moves that limb, the mobile moves as well. Gradually, the child discovers that its own movement has caused the change, establishing one of the fundamental relationships between an action and its consequence.
The FEE CTU researchers replaced the infant with the humanoid robot iCub and created a computational model for it – a simplified “artificial brain.” The robot was not told in advance which limb was connected to the mobile or exactly how its movements would affect the environment. It had to discover these relationships gradually through its own movements and their consequences.
No Discovery Without Surprise – for Robots or Children
Expectations and curiosity play a key role in the model. The robot continuously generates predictions about what will happen when it moves a limb in a particular way. If reality differs from what it expected, a “surprise” occurs. It is precisely this unexpected outcome that the model finds interesting and that motivates further exploration. Once the robot learns to predict a particular relationship correctly, it is no longer novel and its attention can shift elsewhere.
“A child has to learn which changes in its environment are caused by its own actions. Our model works with basic cognitive mechanisms – prediction, surprise and curiosity. The robot expects a certain consequence of its movement. When something unexpected happens, it becomes interesting to the robot, and it tries similar movements again. In this way, it gradually learns the relationship between its own body and the surrounding world,” explains Associate Professor Matěj Hoffmann, head of the Humanoid Robotics Laboratory at the Department of Cybernetics, FEE CTU.
One advantage of using a robot as a model is that researchers know exactly what its “brain” contains and can deliberately modify its individual mechanisms. In so-called ablation experiments, the researchers therefore switched off parts of the model one by one and observed whether the robot was still able to understand the relationship between its own movement and the response of the object it was observing. The results showed that prediction and exploration mechanisms are important for the robot’s behaviour, as is, for example, a certain degree of motor noise.
It Is Not Enough to See Which Arm Moves More
The results also offer a new perspective on how classic experiments with infants are interpreted. These studies often assume that once a child understands which limb controls the suspended mobile, it will begin to move that particular limb more.
The experiments with iCub, however, revealed a more complex picture. The robot was able to discover the relationship between its movement and the response of the mobile in different ways. In some cases, it did indeed move the connected limb more. In others, however, it moved it less – but in a very specific way that effectively caused the mobile to move. The total amount of movement may therefore not be a sufficient indicator of whether a child has understood the relationship between its action and its consequence.
“The classic interpretation can be simplified as follows: if the connected arm moves more, the child has understood that it controls the mobile. We show, however, that other strategies are possible. The robot sometimes moved the connected arm less, but made precisely those movements that produced an interesting effect. For developmental psychology, this is an important indication that we should look not only at how much a child moves, but also at how it moves,” says Dr. Sergiu T. Popescu, a developmental psychologist working in the FEE CTU research group.
According to the authors, this diversity of strategies may also help explain some of the variability seen in psychological studies. The mobile paradigm has been used for decades, yet its results have not always been replicated consistently. The new robotic model makes it possible to examine in detail not only the resulting behaviour, but also the mechanisms that led to it.
A Robot as a Tool for Understanding the Developing Brain
The research demonstrates one of the key advantages of developmental robotics: researchers can create a model of a particular mechanism of human cognition, embody it in a physical system and then examine its behaviour precisely in the real world. Unlike a purely computer-based simulation, the experiment also includes the actual properties of the robot’s body, motors and sensors, as well as its physical interaction with the environment.
For Hoffmann’s team, humanoid robots are increasingly becoming a tool for studying the early development of children and their brains. The research naturally combines robotics and artificial intelligence with developmental psychology and neuroscience.
In the longer term, this line of research may also be relevant for robotics itself. A robot that is to learn in a way similar to a child must first develop a basic understanding of its own body, determine which sensory experiences are consequences of its own actions, and gradually discover what it is capable of influencing in its surroundings. The current study represents one step along this path.
“For us, the humanoid robot is primarily a tool for understanding early child development. Its advantage is that we have the entire model under control. We can switch individual mechanisms on and off and observe what changes. Of course, we cannot do that with the human brain,” Hoffmann adds.
iCub: a Child-Sized Robot for Studying Cognition
iCub is a humanoid robot developed by the Italian Institute of Technology (IIT) for research into cognition and interaction with the environment. It stands just over one metre tall, roughly the size of a four-year-old child, and its body is controlled by 53 electric motors. It perceives the world through cameras, microphones and thousands of touch sensors embedded in its electronic skin. The Prague-based group from the Department of Cybernetics at FEE CTU acquired iCub as part of the Research Centre for Informatics project.
Article in Science Robotics
Photos and Videos
https://drive.google.com/drive/folders/1ZbdNLyJZWgOXK1fqgMFGd1VkknGU69H5?usp=sharing
