What guidance heuristics should be used to explore vast sensorimotor spaces in unknown changing bodies and environments?
With a sufficiently rich environment and multimodal set of sensors and effectors, the space of possible sensorimotor activities is simply too large to be explored exhaustively in any robot’s life time: it is impossible to learn all possible skills. Moreover, some skills are very basic to learn, some other very complicated, and many of them require the mastery of others in order to be learned. For example, learning to manipulate a piano toy requires first to know how to move one’s hand to reach the piano and how to touch specific parts of the toy with the fingers. And knowing how to move the hand might require to know how to track it visually.
In MACSi, we will adapt existing intrinsic motivation systems, and denoted IAC architectures, to high-dimensional sensorimotor spaces, so that it can guide motor exploration to collect efficiently motor learning examples. Furthermore, we will formalize, implement and evaluate two kinds of combination of advanced intrinsic motivation systems (using IAC) and simple forms of social guidance, to operate in large continuous sensorimotor spaces. The first kind will involve social cheering, allowing the human to provide a scalar one-dimensional feedback on a “permitted-forbidden” scale, which will allow the human to encourage the robot to pursue certain sensorimotor activities and to abandon some others. The second kind will involve stimulus enhancement, which consists in drawing the attention of the learner to a particular object in the environment, where he then learns on its own using intrinsically motivated exploration. We expect that social guidance will significantly speed up the exploration of sensorimotor spaces, while intrinsic motivation will allow robot exploration and learning to require much less effort and involvement from the human.
Challenge leader : FLOWERS team at INRIA Bordeaux (Bordeaux).
Dernière modification le 31/05/2011