Abstract:Developmental
robotics is an emerging field that blends the boundaries between robotics,
artificial intelligence, developmental psychology, and philosophy. The basic
research hypothesis of developmental robotics is that truly intelligent robot
behavior cannot be achieved in the absence of a prolonged interaction with a
physical or a social environment. In other words, robots must undergo a
developmental period similar to that of humans and animals. Using a new
paradigm based on observations of how human infants develop their skills, the goal
of this field is to create autonomous robots that are more intelligent, more
adaptable, and more useful than the robots of today, which can only function in
very limited domains and situations.
This talk
will focus on recent research results that show how a robot can solve multiple
tasks based on what it learns during a developmental period similar to a
child’s play. During this period the robot actively tries to grasp, lift,
shake, touch, scratch, tap, push, drop, and crush objects. At the end of this
period the robot knows what different objects sound like when they are dropped,
feel like when they are squeezed, etc. Because these properties are grounded in
the robot’s sensorimotor repertoire the robot can autonomously learn, test, and
verify its own representations without human intervention. The talk will
demonstrate how the robot can use this information to recognize objects,
separate objects into functional categories, and even find the odd-one-out in a
set of objects. The talk also will demonstrate how the robot can use
sensorimotor interactions to bootstrap the development of its visual system in
the context of a button-pressing task.
Results and
videos will be presented for two different humanoid platforms.