Can Reinforcement Learning for Continuous Control Generalize Across Physics Engines?

TL;DR

This paper investigates whether reinforcement learning algorithms trained in one physics engine can generalize to others, revealing that transferability varies significantly across engines and algorithms, with MuJoCo showing better generalization than PyBullet.

Contribution

The study provides a comparative analysis of RL generalization across different physics engines, highlighting the conditions under which transferability is feasible.

Findings

MuJoCo enables better transfer to other engines.

RL algorithms do not generalize well from PyBullet.

Reducing seed variability improves algorithm generalizability.

Abstract

Reinforcement learning (RL) algorithms should learn as much as possible about the environment but not the properties of the physics engines that generate the environment. There are multiple algorithms that solve the task in a physics engine based environment but there is no work done so far to understand if the RL algorithms can generalize across physics engines. In this work, we compare the generalization performance of various deep reinforcement learning algorithms on a variety of control tasks. Our results show that MuJoCo is the best engine to transfer the learning to other engines. On the other hand, none of the algorithms generalize when trained on PyBullet. We also found out that various algorithms have a promising generalizability if the effect of random seeds can be minimized on their performance.