Learning controllers for human-robot interaction

摘要: In order for robots to assist and interact with humans, they must be socially intelligent. Social intelligence is the ability communicate understand meaning through social interaction. Artificial can broadly described, as an effort describe simulate property of human inside a computational model. most cases, this simulation happens in vacuum. An agent, such robot, maintains model which includes all information required make decisions. This internal representation what we consider its intellect. Information may enter perception, expressed form action. separation knowing doing quite effective representing certain types intelligence. However it does not lend itself simulating cognition. In socially, agent able affect change mental other agents, well physical world. However, interaction creation inherently different than interactions Because there are no mathematical models how actions perceptions representations human, cannot hope build interactive by directly process. For reason, when building artificial intelligence, need pay attention prevailing theories on humans learn. Many popular from cognitive science, psychology language development suggest that action perception subordinate representations. Instead, result results agent's environment agents. particular, learning theory says process allows agents one another ground up, starting resulting shared representations, understanding others. This thesis addresses problem into robotic systems using adaptive control. We focus use decision theoretic planning bottom up. first examine recognition designing human-friendly control strategies. Next, address defining subjective measures interactivity leveraging expertise. Finally, define evaluate method participating situated emphasize predict modulate observable responses rather attempting infer their or emotional states. The effectiveness demonstrated experimentally custom systems.