How does a young child learn that the movement of his hand can change the world around him? And what is happening in his brain? An international team of scientists from the Faculty of Electrical Engineering, CTU (Czechia), Universität Hamburg and Université Paris Cité searched for answers using the humanoid robot iCub.
The results of the experiments, which provide a new insight into the mechanisms of early learning, are now published in Science Robotics, the most prestigious scientific journal in the field of robotics.
Scientists from the Faculty of Electrical Engineering at CTU used the humanoid robot iCub as a model for studying one of the basic abilities that we acquire in the first months of life: the understanding that we can influence the world around us with our own movements.
The experiment connects robotics and artificial intelligence with developmental psychology and neuroscience. Its main goal is not to teach the robot a new practical task, but to better understand the mechanisms underlying the early development of human cognition using a robotic model.
The team of associate professor Matěj Hoffmann from the Department of Cybernetics at CTU FEL followed up on the classic research of American psychologist Carolyn Rovee-Collier in the experiment. In the experiment, called the “mobile paradigm,” a child lies in a crib and one of its limbs is connected to a hanging toy. When the child moves the limb, the toy moves.
The child gradually discovers that it is his own movement that caused the change – thus creating one of the basic relationships between his own action and its consequence.
Researchers from the Faculty of Electrical Engineering and Computer Science of the Czech Technical University replaced the child with the humanoid robot iCub and created a computational model for it – a simplified “artificial brain”.
The robot did not receive any information in advance about which limb was connected to the toy or how exactly its movement would affect the surroundings. It had to gradually come to these connections through its own movements and their consequences.
There is no discovery without surprises. For a robot or a child
Expectations and curiosity play a key role in the model. The robot continuously makes predictions about what will happen when it moves a limb in a certain way. If reality differs from expectations, a “surprise” arises. It is the unforeseen result that is interesting to the model and motivates it to further investigate. Once the robot learns to correctly predict a given relationship, it ceases to be new and its attention can shift elsewhere.
A child needs to learn which changes in their environment are caused by their 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 finds it interesting and tries to make similar movements again. This is how it gradually learns the relationship between its body and the surrounding world.
Dr. Matěj Hoffmann, head of the Humanoid Robotics Laboratory at the Department of Cybernetics at the Faculty of Engineering and Technology of Czech Technical University.
The advantage of a robot as a model is that scientists know exactly what its “brain” contains and can change its individual mechanisms in a targeted manner. In so-called ablation experiments, the researchers therefore gradually switched off parts of the model and monitored whether the robot was still able to understand the relationship between its movement and the reaction of the observed object. The results showed that prediction and exploration mechanisms are important for its behavior, but also, for example, a certain level of motor noise.
It’s not enough to just watch which hand moves more
The results also provide a new perspective on the interpretation of classic experiments with children. These are often based on the assumption that if a child understands which limb controls a suspended toy, he will start moving that limb more.
However, the experiments with the iCube showed a more complex picture. The robot was able to discover the relationship between movement and the reaction of the toy in different ways. In some cases, it actually moved the attached limb more, but in other cases, it moved it less – but in a very specific way that effectively caused the toy to move.
Therefore, the amount of movement alone may not be a sufficient indicator of whether the child understood the relationship between its action and its consequence.
“The classic interpretation says in simple terms: when the attached hand moves more, the child understood that he was controlling the toy. But we show that there are other strategies. The robot sometimes moved the attached hand less, but made exactly those movements with it that had an interesting effect. For developmental psychology, this is an important stimulus that it is necessary to monitor not only how much the child moves, but also in what way,” says Dr. Sergiu T. Popescu, a developmental psychologist working in the research group of the Faculty of Electrical Engineering, Czech Technical University.
According to the authors, it is precisely the diversity of strategies that can help explain part of the variability in psychological studies. The mobile paradigm has been used for decades, but its results have not always been clearly reproduced. The new robotic model allows for detailed monitoring not only of the resulting behavior, but also of the mechanisms that led to it.
Robot as a way to understand the child’s brain
The research shows one of the advantages of so-called developmental robotics: scientists can create a model of a certain mechanism of human cognition, put it in a physical body and then precisely study its behavior in the real world. In contrast to a purely computer simulation, the actual properties of the robot’s body, its motors, sensors and physical connection to the environment also enter the experiment.
For Hoffmann’s team, humanoid robots are increasingly a tool for studying the early development of a child and his brain. The research naturally connects robotics and artificial intelligence with developmental psychology and neuroscience.
In the longer term, this direction may also have significance for robotics itself. A robot that is supposed to learn in a similar way to a child he must first gain a basic understanding of his own body, discover which sensations are the result of his own actions, and gradually discover what he can influence in his surroundings. The current study represents one of the partial steps on this journey.
“For us, the humanoid robot is primarily a tool for understanding the early development of a child in this research. Its advantage is that we have its entire model under control. We can turn individual mechanisms on and off and observe what changes. Of course, we cannot do that with the human brain,” adds Hoffmann.
The iCub robot: a child among robots
The iCub is a humanoid robot designed by the Italian Institute of Technology (IIT) for research into cognition and interaction with the environment. It measures just over one meter, is about the size of a four-year-old child, and its body controls 53 electric motors. It perceives the world through cameras, microphones, and thousands of touch sensors built into its electronic skin.
The Prague group from the Department of Cybernetics at the Faculty of Electrical Engineering and Mechanical Engineering of the Czech Technical University acquired the iCub as part of a project by the Research Center for Informatics.
