A few words about my work...

Here we are! This is a brief introduction on what is my work about. As you might have already guessed I'm up on something related to robotics. My research stems from an interdisciplinary approach ranging from "brain sciences" to actual robotics (or AI perhaps). I'm interested both on what biology could tell us about complex systems and what robotics could tell us about ourselves. I've addressed these issues in my Ph.D. thesis although in a simplified form. I've found some of the topics intriguing. My long-term goal (project) is to build a complete alive artificial system (whatever this means) embedded in a real body, learning, growing (in some sense), feeling, perceiving and acting, naturally interacting with us. Wouldn't it be great? Along the way perhaps we might learn how we are made, we might get insights on our "physiology".

In order to investigate the many issues involved and provide sensible replies to the many pending questions I've studied "sensorimotor development" as a model of learning and adaptation from the neural sciences and robotics perspective. I realized though that this is not enough. A more global point of view is needed. It's perhaps far too naive to say that a complex system is not simply the sum of its parts. Cognition, feelings, attention, value systems, complex multisensory integration, learning, speech and many other aspects are needed. Development, life-like growth, I argue embeds all these and thus could provide the glue and a possible schema where to frame things coherently. 

Development is the process that governs the time-varying nature of the learning agent itself. Consequently, on one hand the analysis of developmental processes might provide unique insights on how sensorimotor coordination arises in biological learners, on the other it could be the only feasible procedure to design highly complicated artificial systems.

Of course because to convince the audience we have to show something after all, I built a "first prototype" of a "baby humanoid robot" named Babybot. I demonstrated how the twelve degrees of freedom "baby" acquires autonomously orienting and reaching behaviors, and what is the advantage of the proposed framework in correctly weighing the various constraints. 

In our crude artificial implementation we've shown that simple initial behaviors can be seen as the building blocks, guiding the learning of more sophisticate behaviors, and acting as a bootstrap procedure for the whole system. At the same time, the simple behaviors can also serve to keep subsequent learning processes within feasible regions of the "space state". 

One key point of the whole procedure is complexity control. This concept borrowed from statistical learning theory tells us that the goal of a developing system might be seen as that of balancing the complexity of its internal learning structures to that of its experience thus allowing optimal learning to take place.

This is of course work in progress, I've such a nice plan on what to do next, stay tuned for more news!

"One day they'll be like us, undistinguishable, alive, feeling... isn't this a nice dream?"