|
|
 |
 |
|
|
|
| How to satisfy a curious robot? / Frédéric Kaplan |
|
| In November 2005, more than thirty people – most of them designer students from ECAL (Ecole Cantonale d’Art de Lausanne) – joined me in Lausanne to take on the following challenge: to satisfy the whims of a curious robot. Our ultimate goal was to build a playroom where a robot could engage in various new learning activities. |
|
After a few days of intense exploration, all our efforts converged on a set of smaller objects, toys and infrastructures that were chosen for their originality, feasibility and their scientific interest. These included a “universal gripper” with which the robot can pick up everyday objects, a special pen that introduces the robot to the joys of drawing, a switch with which it can control electrical devices, a mirror that takes into account the specific morphology of the robot’s head and the position of its sensors, various toys that jump or can be pushed or pulled, a rocking chair to take a nap, a raincoat, and even a children’s bike. None of these objects contained any technology, but it is incredible how much they enlarged the robot’s range of possible activities.
This project was actually part of years of research into cursiosity-driven robots that my colleague Pierre-Yves Oudeyer and myself have conducted at Sony CSL. Very much like a child can be fascinated by objects in its environment and have great fun by just picking them up, sticking them into their mouths, or rattling them, we are interested in how we can turn robots into curious creatures. We have therefore developed algorithms that make robots eager to explore their environment and choose among different situations those that are more likely to lead to progress in learning. As a consequence, the robot gets bored with situations that are too familiar and avoids activities that are still too complex for its level of development. In order to make progress, such a robot should thus be placed in a stimulating environment that is adapted to its body, its learning biases and its capacities of perception and action.
The idea of working with designers presented itself when I was attending a workshop in the small Swiss town of Chateau-d’Oex, organized by the CRAFT team from the Swiss Federal Institute of Technology (EPFL). This workshop brought together architects, designers and researchers in artificial intelligence. I had a discussion with Christophe Guignard, professor at ECAL, about our work in Paris on artificial curiosity. I explained to him how we were aiming to create objects that stimulated robot learning but that our first attempts were kind of clumsy. For him, that was a job for designers and he suggested to organize a special week dedicated to the construction of a robot’s playroom, later in the year.
This is how I found myself in Lausanne for what proved to be an incredibly intense, but also very rewarding week of work. For supervising the students, I teamed up with Martino d’Esposito, a young and talented industrial designer who teaches at ECAL and designs objects and furniture for many famous companies. While he was helping the students to finalize their first prototypes, I was busy adapting the robot’s behaviours to see whether it would be able to learn how to use these new toys. After a week only, the robot was able to explore a multitude of new interactions: drawing, rolling, riding a bike, pulling light or heavy objects, seeing itself in the mirror, playing new games, and even wearing new clothes. These objects increase the range of action of the machine, like extensions of its body. Its world becomes more open and challenging, and in certain ways a little closer to our own.
Intelligence cannot be explained solely by the organization of our brain. It results from a long history of “embodied interactions” in rich environments. It takes long hours of discovery, exploration and investigation for gradually learning how to perceive and act. Designing a robot’s playroom was for us an exhilarating scientific and artistic adventure. Hopefully, our robots will learn a lot with this playroom. We sure did.
Frederic Kaplan.
|
|
|
|
|
|
|
|
|
 |
|
|
