Bayesian cognitive models for imitation

Imitation offers a powerful mechanism for knowledge acquisition. Before human infants acquire language, much of their knowledge is gained via imitation. Increasingly, researchers hope that endowing machines with imitative abilities will bestow the flexibility, generality, and social awareness of humans. This dissertation illustrates how the interplay between cognitive studies and robotics can yield insights into both fields. Specifically, I develop several algorithms for imitation learning based on constraints suggested by cognitive results in humans. Results include simulations and implementation on a humanoid robot. I also propose a model of human social interaction based on studies of infants conducted using a humanoid robot. Given the increasing importance of probabilistic reasoning with prior knowledge in machine learning, and the mounting evidence that humans employ Bayesian models for decision making, my results are unified in a Bayesian formalism.