A Review of Nine Physics Engines for Reinforcement Learning Research

TL;DR

This paper addresses the challenge of selecting appropriate physics engines for reinforcement learning by conducting a systematic, two-pronged review of nine engines (Brax, Chrono, Gazebo, MuJoCo, ODE, PhysX, PyBullet, Webots, and Unity). It combines a popularity analysis based on citation data with a feature-focused assessment of documentation, ecosystem support, and MARL capabilities to map strengths and trade-offs. Key findings identify MuJoCo as the dominant engine for performance and flexibility, while Unity offers superior usability at the expense of scalability and fidelity, and Brax presents a GPU-accelerated but still maturing option. The study emphasizes transparency, reproducibility, and the need for better cross-engine benchmarking to accelerate progress in RL research and tool development.

Abstract

We present a review of popular simulation engines and frameworks used in reinforcement learning (RL) research, aiming to guide researchers in selecting tools for creating simulated physical environments for RL and training setups. It evaluates nine frameworks (Brax, Chrono, Gazebo, MuJoCo, ODE, PhysX, PyBullet, Webots, and Unity) based on their popularity, feature range, quality, usability, and RL capabilities. We highlight the challenges in selecting and utilizing physics engines for RL research, including the need for detailed comparisons and an understanding of each framework's capabilities. Key findings indicate MuJoCo as the leading framework due to its performance and flexibility, despite usability challenges. Unity is noted for its ease of use but lacks scalability and simulation fidelity. The study calls for further development to improve simulation engines' usability and performance and stresses the importance of transparency and reproducibility in RL research. This review contributes to the RL community by offering insights into the selection process for simulation engines, facilitating informed decision-making.

Figures (2)

• Figure 1: Two similar environments realized in different engines
• Figure 2: Yearly citations of the frameworks' original publications