Der Roboter iCub lernt laufen: 5 Fragen an Giorgio Metta

Durch Versuch und Irrtum erarbeitet er sich Strategien, um seine Aufgaben erfüllen zu können. Die Rede ist von dem iCub Robot. Er lernt durch Beobachten und Nachahmen von Menschen. Die Roboter-Software wird als Open-Source-Software entwickelt. Giorgio Metta, der das Projekt iCub leitet, hat meine Fragen beantwortet.  (Das Interview ist in englischer Sprache  und ist urheberrechtlich geschützt). 

iCub is the most popular humanoid robot in the world. In what kind of fields can it be employed? What are the advantages of employing it?

G.M.: iCub is a research platform and as such the only application is in fact research. It’s a very flexible robot with many sensors and sophisticated kinematics to support research ranging from motor control, balancing, walking up to computer vision, object manipulation, visuotactile integration, and human-robot interaction. Many robots are very specialized (e.g. walking) and cannot flexibly support complex tasks (e.g. mobile manipulation) as the iCub can do.

What is the most innovative step achieved in the technical development of iCub?

G.M.: From the hardware point of view, we developed a whole-body “dense” artificial skin which I believe is unique. On the software side, force control and consequently whole-body control in the presence of multiple contacts with the environment.

What did iCub learn in the last years? How autonomous is its interaction and cooperation with humans?

G.M.: iCub recently learned to walk. I must confess that the iCub is not completely autonomous during an interaction. We’re not ready yet for a full-fledged human-robot interaction. We’re working on it.

iCub was used to support the investigation of the embodied mind thesis and to test the hypothesis behind embodiment. What kind of results were reached in the investigation of embodied cognition in the interaction with the environment?

G.M.: In general, we’ve been able to show that AI algorithms tailor-made for the robot work better, faster, more accurately. This is because we can exploit the robot’s actions to acquire data and process it in real time and guarantee that training and exploitation of the leant skills share the same epistemology (you learn in the same context where you will apply the acquired skills). This circumvents the poor (typically) generalization ability of present-day AI.

How do people react to iCub when they experience it for the first time?  

G.M.: In many different ways. On average, people are astonished by the movement skills of iCub and clearly by its child-like appearance.

Unsere Updates zum Thema "Innovationen wirksam kommunizieren" nur einen Klick entfernt!
I agree to have my personal information transfered to MailChimp ( more information )