If robots had cognition—if they could memorize, learn, understand, adapt, and reason—we could teach them to play our favorite games, set the dinner table, or help us with other chores around the house. With cognition, robots could be useful in all kinds of environments, including our homes, workplaces, public places like museums and airports, and even schools. It turns out, however, that it’s really hard to make a cognitive robot. Think about it: to set your dinner table, a robot would need to know how to move its body, how to recognize plates, cups and utensils, how to pick these things up and put them down, where each plate goes, and what to do if you have a guest over for dinner. Pre-programming a robot with all these skills in advance is nearly impossible, especially since everyone has different plates, different homes, and different dining habits. It would be a lot of work! And what if you wanted to program your robot to do more than just set the table? What if you wanted it to also empty your dishwasher, make your bed, help you with homework, or take the garbage out? You can’t possibly pre-program a robot with all the information it needs to do these tasks . . . but what if robots could develop these skills over time—just like how people learn to do new things? If robots could learn like humans learn, they could adapt to all kinds of environments, and develop the skills needed to help us with all kinds of different jobs! But how can you get a robot to learn? It turns out that researchers who study human development—also called developmental psychologists—are working together with robotics researchers to try and answer this very question. Humans are born with very few skills. When we are babies, we can hardly control our arms, legs and neck muscles, but within just a few years we can roll, sit, crawl, grasp objects, walk, and talk. Before we know it, we have mastered enough skills to do more complicated tasks, like setting the dinner table, or even learning a new language! How do young children learn new tasks, like catching a ball or balancing? How do they learn new ideas? How do they improve their skills over time? These are questions that brain and development experts are studying, and they are questions that robotics researchers are studying as well. They want to know: if we can better understand how humans learn, will we be able to make cognitive robots that can learn over time too? In 2004, researchers in Europe decided to try and answer this very question by building a humanoid robot that is about the size of a small child. The project was called RobotCub and they called their baby robot the iCub. The iCub is 1 meter (3.3 feet) tall and weighs 22 kilograms (48.5 pounds) . . . about the size of an average preschool-aged child. The iCub has a head, eyes, eyelids, lips, arms, hands, a waist, and legs—just like a person. But instead of muscles, the iCub uses 53 motors to move all those parts. The iCub also needs to know what’s happening around it, so it has all kinds of sensors to help it understand its environment and its body position. For example, it has two cameras to see, two microphones to hear, and smart “skin” that lets it feel if someone is touching it. It even has sensors on all of its joints (such as its elbows, knees, and hips) so that it can know where all its body parts are and how they are moving. For example, a sensor on the iCub’s elbow might let the robot know if its arm is stretched out or bent. The iCub also has accelerometers and gyroscopes inside its head (they are called “inertial sensors”) that tell it how fast it is moving or turning, and give iCub the sense of balance. You also have a pair of inertial sensors called “vestibular systems” located inside your head in the inner part of your ears. That’s how you can balance or walk in the dark, and how you feel the acceleration when you ride a car or an elevator. In all, the iCub has more than 5,000 pieces! Why does the iCub look the way it does? Researchers believe that the shape of our body affects the way we develop new skills. For example the shape of your feet and hands allows you to walk on two legs and to eat an ice cream cone at the same time. Can you imagine a cat or a horse doing this? Walking while holding an ice cream cone—and doing this without falling—are some of the “cognitive skills” we develop as we grow up. The idea that a brain develops within a body that can interact with the world is called embodied intelligence—which literally means “intelligence in a body.” Embodied intelligence is what makes a robot different from a computer that can’t explore and interact with the physical world. A robot can look for an ice-cream cone and take it to you while a computer can only show pictures . . . and you know the difference between the picture and the real thing! If you’re trying to build a robot that can learn the way a young child learns, it makes sense to give it a child-like body so that it can learn similar skills. That’s why the iCub looks like a small child. Having a human-like body makes it easier for people to teach the iCub new skills. If you want to teach a baby to pick up a ball, you might guide the baby’s hand to the ball, and show him how to squeeze it with his fingers, or you might show him how by picking it up yourself, and then encourage him to do the same. But imagine trying to teach a dishwasher how to pick up a ball—it would be hard, because a dishwasher doesn’t have the same body parts that you have. Because the iCub has body parts that are similar to ours, we can simply show it how to do new skills—and this is much easier than having to write a computer program to tell the robot what to do! For example, to teach the iCub to pick up a ball, you could point to the ball, wait until the iCub nods its head to show that it’s learned what the ball is, and then show the iCub how to pick it up and what to do with it. Because the iCub has arms and hands that are similar to yours, it will be able to imitate what you are doing. The iCub is also very gentle, so you can also take its arm and guide it to show it special skills. By turning its head to look around, nodding to show understanding, or asking questions, the iCub can also show that it understands you. Using its eyes, eyelids, eyebrows, mouth and body motion, the iCub can be very expressive. There is a whole field of study that looks at how humans and robots interact, called “human-robot interaction.” This kind of research is really important if everyday people (and not just robot experts) are going to work or play with robots. Check out the video above. Can you tell what the robot is “thinking?” It also turns out that the iCub can help researchers understand human cognition. It’s very hard to study child development because it takes time for a child to grow, and researchers are not allowed to interfere with the natural process of a baby growing up. By programming the iCub to learn new skills, and observing how it behaves, psychologists and child development researchers (with a bit of help from robotics engineers) might learn something about how babies develop. What can the iCub do? Designing a cognitive robot is such a huge task that hundreds of researchers from many different disciplines (including mechanical and electronics engineering, computer science, artificial intelligence, mathematics, physiology, and psychology) have had to work together to make progress on it. In fact, more than 25 laboratories around the world now have an iCub! Together, they’ve shown that the iCub can learn to crawl, sit and balance, control its legs and arms, reach for objects, react to touch, discover, memorize and understand how to use objects, and imitate humans. It’s even learned how to play the drums! One thing that has helped the iCub project succeed is that it is an open-source platform, meaning all the hardware and software developed for the project is free and available for anyone to use. This is really helpful, since researchers can easily share their work and learn from each other. What can't it do (yet)? The iCub doesn’t have its own energy supply, or computers that are powerful enough to run all of its software. That’s why the robot is attached to a cable that powers it and sends it commands from an outside computer. Without this cable, the robot wouldn’t be able to operate. The iCub was also built for a laboratory environment, which tends to be more controlled and predictable than the real world, which can be messy. Because of this, the iCub wouldn’t work very well in your home, or outside. A new model with batteries and wireless capabilities is being built to allow the iCub to exit the laboratory. You can also see from the video that the iCub still has much to learn; it might be 10 years old, but it still can’t do many of the things that a human baby can do. It will take researchers many, many years before that happens. Hopefully, iCub’s open-source platform will help researchers share what they have learned and keep its development moving forward.