Learning a set of interrelated tasks by using a succession of motor policies for a socially guided intrinsically motivated learner


Nov 03, 2017 (modified: Nov 03, 2017) ICLR 2018 Conference Blind Submission readers: everyone Show Bibtex
  • Abstract: We propose an active learning architecture, capable of organizing its learning process to learn complex motor policies (which are succession of primitive motor policies) achieving multiple outcomes: Socially Guided Intrinsic Motivation at High Level (SGIM-HL). The learner can generalize over its experience to continuously learn new outcomes, by choosing actively what and how to learn guided by empirical measures of its own progress. In this paper, we are considering the learning of a set of interrelated complex outcomes hierarchically organized. We introduce a new framework called "procedures", which enables the autonomous discovery of how to combine previously learned skills in order to learn increasingly more complex motor policies (combinations of primitive motor policies). Our architecture can actively decide which outcome to focus on and which exploration strategy to apply. Those strategies could be autonomous exploration, or active social guidance, where it relies on the expertise of a human teacher providing demonstrations at the learner's request. We show on a simulated environment that our new architecture is capable of tackling the learning of complex motor policies, to adapt the complexity of its policies to the task at hand. We also show that our "procedures" increases the agent's capability to learn complex tasks.
  • TL;DR: The paper describes a strategic intrinsically motivated learning algorithm which tackles the learning of complex motor policies.
  • Keywords: developmental robotics, intrinsic motivation, strategic learning, complex motor policies