In open-ended environments, autonomous learning agents must set their own
goals and build their own curriculum through an intrinsically motivated
exploration. They may consider a large diversity of goals, aiming to discover
what is controllable in their environments, and what is not. Because some goals
might prove easy and some impossible, agents must actively select which goal to
practice at any moment, to maximize their overall mastery on the set of
learnable goals. This paper proposes CURIOUS, an algorithm that leverages 1) a
modular Universal Value Function Approximator with hindsight learning to
achieve a diversity of goals of different kinds within a unique policy and 2)
an automated curriculum learning mechanism that biases the attention of the
agent towards goals maximizing the absolute learning progress. Agents focus
sequentially on goals of increasing complexity, and focus back on goals that
are being forgotten. Experiments conducted in a new modular-goal robotic
environment show the resulting developmental self-organization of a learning
curriculum, and demonstrate properties of robustness to distracting goals,
forgetting and changes in body properties.

该论文提出了一种名为 CURIOUS 的算法，它利用 MODULAR Universal Value Function Approximator 和自动化课程学习机制来实现学习代理的自主目标设定和自我组织学习课程，实现学习目标的快速最优化。