This images, inspired on Gibson's visual space, shows that the same
gesture can have an huge difference in appearance. When doing imitation
this visual perspective must be taken into account.
Image idea Manuel Lopes, Design Francisco Melo
Baltazar grasps an object. Grasping is a very complex skill, it
requires localizing the object, choose a good grasping point, plan a
reaching behavior, execute the reaching and final close the hand.
Baltazar likes to play with objects. In this case it is executing a
task that was learned from a demonstration.
Baltazar likes carrots!!!
Robots are very complex machines. This is the team that design and
iCub can localize sounds. This is how it does it.
Baltazar likes to imitate. Here it copies only the observed effects
without taking into account the demonstrator.
iCub likes to know everything that goes around him. An egosphere
includes information about the motion, color, sound and edges that are
When imitating we do not consider everything. When saying goodbye do
you use the same elbow motion? Or just the hand motion?
How can you learn a relation where a single input can have more than
Policy gradient methods are a good choice to improve control policies.
How can you reuse all your data?
Gesture imitation is easy. Or maybe not...
One of the first pictures of the iCub head.
One of the first pictures of the iCub head before being built!
iCub is tested as a real child.
iCub with very nice ears. Symmetrical ears are bad!
Sometimes we need a good wizard of Oz.
Imitation is not just for learning. It is also for social acceptance.
A mosaic of the space around iCub.
What is the stationary point of two learners that do not know that they
are playing a game?
A parallel mechanism was also studied as a potential iCub neck.
Baltazar showing its skills.
Baltazar design phase.