Intrinsically Motivated Learning of Hierarchical Collections of Skills

摘要: Humans and other animals often engage in activities for their own sakes rather than as steps toward solving practical problems. Psychologists call these intrinsically motivated behaviors. What we learn during behavior is essential our development competent autonomous entities able to efficiently solve a wide range of problems they arise. In this paper present initial results from computational study learning aimed at allowing artificial agents construct extend hierarchies reusable skills that are needed autonomy. At the core model recent theoretical algorithmic advances reinforcement learning, specifically, new concepts related algorithms with skill hierarchies.