Asynchronous Reinforcement Learning for Real-Time Control of Physical Robots

TL;DR

This paper demonstrates that asynchronous reinforcement learning enables real-time control of physical robots more effectively than sequential methods, especially when learning updates are computationally expensive, leading to faster, more responsive robotic behaviors.

Contribution

The authors systematically compare sequential and asynchronous reinforcement learning on real robots, showing asynchronous methods maintain responsiveness and outperform sequential ones under costly updates.

Findings

Asynchronous RL maintains appropriate action cycle times under high update costs.

Sequential RL performance degrades with increased learning update times.

The system learns to reach and track visual targets from pixels within two hours on real robots.

Abstract

An oft-ignored challenge of real-world reinforcement learning is that the real world does not pause when agents make learning updates. As standard simulated environments do not address this real-time aspect of learning, most available implementations of RL algorithms process environment interactions and learning updates sequentially. As a consequence, when such implementations are deployed in the real world, they may make decisions based on significantly delayed observations and not act responsively. Asynchronous learning has been proposed to solve this issue, but no systematic comparison between sequential and asynchronous reinforcement learning was conducted using real-world environments. In this work, we set up two vision-based tasks with a robotic arm, implement an asynchronous learning system that extends a previous architecture, and compare sequential and asynchronous reinforcement learning across different action cycle times, sensory data dimensions, and mini-batch sizes. Our experiments show that when the time cost of learning updates increases, the action cycle time in sequential implementation could grow excessively long, while the asynchronous implementation can always maintain an appropriate action cycle time. Consequently, when learning updates are expensive, the performance of sequential learning diminishes and is outperformed by asynchronous learning by a substantial margin. Our system learns in real-time to reach and track visual targets from pixels within two hours of experience and does so directly using real robots, learning completely from scratch.