The ability to interpret demonstrations from the perspective of the teacher plays a critical role in human learning. Robotic systems that aim to learn effectively from human teachers must similarly be able to engage in perspective taking. We present an integrated architecture wherein the robot’s cognitive functionality is organized around the ability to understand the environment from the perspective of a social partner as well as its own. The performance of this architecture on a set of learning tasks is evaluated against human data derived from a novel study examining the importance of perspective taking in human learning. Perspective taking, both in humans and in our architecture, focuses the agent’s attention on the subset of the problem space that is important to the teacher. This constrained attention allows the agent to overcome ambiguity and incompleteness that can often be present in human demonstrations and thus learn what the teacher intends to teach.

Many common types of language understanding depend on situational context. In the extreme, utterances like ”the one to the left of the green ones”, ”let’s do that again ” or ”can you help me? ” provide little content or restrictions through their words, but can be readily understood and acted upon by a human listener embedded in the same situation as the speaker. We describe a series of computational models of situated language understanding that take into account the context provided by a game the language users are playing. Starting with a game focusing on spatial disambiguation, we proceed to a model taking into account player’s recognized intentions to perform referent disambiguation and end with a system that understands highly situated commands directly in terms of recognized plan fragments. Finally, we discuss our use of these models in building artificial agents that plan alongside the player in the game world and co-operate through language and their own initiative.

Abstract not found