Meta Learning Shared Hierarchies

TL;DR

This paper introduces a meta-learning approach for hierarchical policy learning using shared primitives, enhancing sample efficiency and transferability across diverse tasks and robotic platforms.

Contribution

It proposes a concrete metric for hierarchy strength and an end-to-end algorithm that leverages off-the-shelf RL methods for learning shared primitives.

Findings

Discovered motor primitives for four-legged robots in maze environments.

Transferred primitives to long-timescale obstacle courses.

Enabled humanoid robots to walk and crawl with the same policy.

Abstract

We develop a metalearning approach for learning hierarchically structured policies, improving sample efficiency on unseen tasks through the use of shared primitives---policies that are executed for large numbers of timesteps. Specifically, a set of primitives are shared within a distribution of tasks, and are switched between by task-specific policies. We provide a concrete metric for measuring the strength of such hierarchies, leading to an optimization problem for quickly reaching high reward on unseen tasks. We then present an algorithm to solve this problem end-to-end through the use of any off-the-shelf reinforcement learning method, by repeatedly sampling new tasks and resetting task-specific policies. We successfully discover meaningful motor primitives for the directional movement of four-legged robots, solely by interacting with distributions of mazes. We also demonstrate the transferability of primitives to solve long-timescale sparse-reward obstacle courses, and we enable 3D humanoid robots to robustly walk and crawl with the same policy.