Benchmarking Full-Stack ROS 2 Simulation Platforms for Mobile Robots
Ángel Soriano, Román Navarro, Juan Manuel Navarrete, Robert Vasquez, Rafael Martín, Carlos Gascón, Joel Ramos
Abstract
Following the official end-of-life of ROS 1 in May 2025, the robotics community is rapidly migrating to ROS 2, necessitating a rigorous re-evaluation of available simulation tools. This paper presents a comparative benchmarking of five widely used simulators: Gazebo (Harmonic), Webots (R2025a), Unity (6.1), NVIDIA Isaac Sim (4.5.0), and Open 3D Engine (O3DE). Using a standardized high-performance workstation and a unified mobile robot model equipped with stereo cameras, 3D LiDAR, and IMU, each platform is evaluated across distinct environments. Quantitative metrics reported include Startup Time, Real-Time Factor (RTF), and computational footprint (CPU, RAM, and GPU usage). Results reveal a clear trade-off between efficiency and visual fidelity: “ROS 2-first” engines like Webots and Gazebo demonstrate competitive startup times and low resource consumption, making them suitable for iterative development and CI/CD pipelines. Conversely, “Graphics-first” engines like Isaac Sim and Unity offer photorealistic rendering suitable for perception training, but at a significantly higher computational cost. Additionally, O3DE is analyzed as a promising native-ROS 2 alternative. To facilitate independent verification and future research, the complete benchmarking suite, including scenes and measurement scripts, is released as an open-source repository.
BibTeX
@inproceedings{ral2026_benchmarkingfull,
title = {Benchmarking Full-Stack ROS 2 Simulation Platforms for Mobile Robots},
author = {Ángel Soriano and Román Navarro and Juan Manuel Navarrete and Robert Vasquez and Rafael Martín and Carlos Gascón and Joel Ramos},
booktitle = {RA-L 2026},
year = {2026}
}