Learning Generalizable Control Programs
This talk will present a framework for guiding autonomous learning in robots. The proposed paradigm allows a robot to organize its sensory and motor resources into hierarchical control programs according to an intrinsic motivation function that finds and models behavioral affordances. The framework encourages behavioral re-use, the acquisition of domain general strategies for interacting with the world, and the efficient generalization of control policies to different contexts. A longitudinal learning experiment on a bimanual robot is provided demonstrating the effectiveness and applicability of the presented framework.