Collective Robot Reinforcement Learning with Distributed Asynchronous Guided Policy Search

TL;DR

This paper introduces a distributed asynchronous Guided Policy Search method enabling multiple robots to collaboratively learn complex manipulation skills, resulting in better generalization and faster training than single-robot approaches.

Contribution

It presents a novel distributed and asynchronous extension of Guided Policy Search for collective robot reinforcement learning.

Findings

Achieves improved generalization across diverse real-world tasks.

Reduces training time compared to single-robot learning.

Demonstrates effective collective learning on a vision-based door opening task.

Abstract

In principle, reinforcement learning and policy search methods can enable robots to learn highly complex and general skills that may allow them to function amid the complexity and diversity of the real world. However, training a policy that generalizes well across a wide range of real-world conditions requires far greater quantity and diversity of experience than is practical to collect with a single robot. Fortunately, it is possible for multiple robots to share their experience with one another, and thereby, learn a policy collectively. In this work, we explore distributed and asynchronous policy learning as a means to achieve generalization and improved training times on challenging, real-world manipulation tasks. We propose a distributed and asynchronous version of Guided Policy Search and use it to demonstrate collective policy learning on a vision-based door opening task using four robots. We show that it achieves better generalization, utilization, and training times than the single robot alternative.