Autonomous Reinforcement Learning via Subgoal Curricula

TL;DR

This paper introduces VaPRL, a method that creates a curriculum of initial states for reinforcement learning, enabling agents to learn complex tasks with minimal human intervention and outperforming existing reset-free RL approaches.

Contribution

The paper presents VaPRL, a novel curriculum-based reinforcement learning approach that reduces environment resets and human interventions while improving learning efficiency and performance.

Findings

VaPRL reduces interventions by three orders of magnitude.

Outperforms prior reset-free RL methods in sample efficiency.

Achieves higher asymptotic performance on robotics tasks.

Abstract

Reinforcement learning (RL) promises to enable autonomous acquisition of complex behaviors for diverse agents. However, the success of current reinforcement learning algorithms is predicated on an often under-emphasised requirement -- each trial needs to start from a fixed initial state distribution. Unfortunately, resetting the environment to its initial state after each trial requires substantial amount of human supervision and extensive instrumentation of the environment which defeats the goal of autonomous acquisition of complex behaviors. In this work, we propose Value-accelerated Persistent Reinforcement Learning (VaPRL), which generates a curriculum of initial states such that the agent can bootstrap on the success of easier tasks to efficiently learn harder tasks. The agent also learns to reach the initial states proposed by the curriculum, minimizing the reliance on human interventions into the learning. We observe that VaPRL reduces the interventions required by three orders of magnitude compared to episodic RL while outperforming prior state-of-the art methods for reset-free RL both in terms of sample efficiency and asymptotic performance on a variety of simulated robotics problems.