Affordances and Emergence of Concepts.

Abstract

In this paper, we used the aordance formal- ization framework (Sahin et al; Adaptive Be- havior, 15(4), 447-472, 2007) to link the con- cepts represented by verbs and nouns in lan- guage to the aordance relations that a robot acquires through its interaction with the en- vironment. First, we argued that a verb is linked not to a specific behavior of the robot, but to a specific eect that dierent behaviors may generate. We showed how demonstra- tions made by a human can allow the robot to learn an eect prototype in its own sensory- motor space, and use this to command the robot to achieve desired goals. Second, we investigated how the aordance relations of objects can be used to create noun concepts (to name objects) and how multi-task learning can utilize the learned object concepts to facil- itate the learning of new aordances. The re- sults are demonstrated on an iCub humanoid robot platform.