ROS 2机器人迁移指南：从Gazebo到Isaac Sim实战

学习如何将ROS 2机器人从Gazebo迁移至NVIDIA Isaac Sim，掌握URDF导入、A/B测试及物理参数调优，确保仿真一致性。

原始标题：Gazebo to Isaac Sim: The Complete ROS 2 Migration

这是一个关于机器人仿真迁移的高级实战课程，核心教你如何将现有的 ROS 2 机器人从 Gazebo（Harmonic） 完美迁移到 NVIDIA Isaac Sim（6.0.1），并通过可重复的数据脚本和 A/B 对比测试，量化证明两者在物理和传感器行为上的一致性。

⚡ 核心学习收获

• 自动化与验证： 编写脚本导入 Gazebo URDF 机器人到 Isaac Sim，并自动反读所有连杆（Links）和关节（Joints）来验证迁移完整性。
• 搭建 A/B 仿真测试平台： 用同一组指令流同时驱动运行在 Gazebo 和 Isaac Sim 中的同一个机器人，通过数值差值量化对比两者的行为。
• 数据驱动的排错能力： 学习通过误差曲线的形状诊断物理引擎差异（例如：水平线代表坐标系偏差、线性增长代表速率差异、陡峭后平直代表遇到极限值、波动代表几何干涉、阶跃代表离散事件）。
• 底层物理参数调优： 深入研究为何 URDF 的阻尼值（Damping）在 PhysX 中会变成速度追踪增益，并修复由此导致的转向不足等经典迁移失败案例。

Published 8/2026
Created by Ferbin Richard
MP4 | Video: h264, 1920×1080 | Audio: AAC, 44.1 KHz, 2 Ch
Level: All Levels | Genre: eLearning | Language: English + subtitle | Duration: 80 Lectures ( 11h 8m ) | Size: 4.8 GB

Migrate a ROS 2 robot from Gazebo to Isaac Sim: URDF import, A/B measurement, PhysX drives, sensors, Nav2 proof

What you’ll learn
⚡ Import a working Gazebo URDF robot into Isaac Sim with a scripted, reproducible importer run, and read back every link and joint to verify what transferred
⚡ Build an A/B rig that drives the same robot in Gazebo and Isaac Sim from one command stream, and prove equivalence with numeric diffs against stated tolerance
⚡ Diagnose a divergence from the shape of its plot: flat means frames, linear means a rate, steep-then-flat means a limit, oscillating geometry, a step an event
⚡ Trace a real migration failure to its mechanism, such as a URDF damping value that becomes PhysX’s velocity tracking gain, and fix it with measured confirmation

Requirements
❗ A working ROS 2 robot in Gazebo, or willingness to build the course’s one from scratch in Section 1. You should be comfortable with a Linux terminal, Python and URDF/xacro.
❗ Ubuntu 24.04 with ROS 2 Jazzy and Gazebo Harmonic. Isaac Sim 6.0.1 needs an NVIDIA RTX GPU; 6 GB VRAM is enough to run both simulators side by side.
❗ No Isaac Sim or Omniverse experience required

Description
This course contains the use of artificial intelligence.

Your robot already works in Gazebo. You have an SDF or Xacro, tuned controllers, and a Nav2 configuration that took weeks to get right. Now you need to move it to NVIDIA Isaac Sim for photoreal sensors, GPU physics, synthetic data, or Omniverse workflows.

The problem is that almost nobody teaches the migration itself.

What this course teaches

You will migrate a working Gazebo robot to Isaac Sim andprove with measurements that it behaves the same, or identify exactly why it does not.

The core of the course is anA/B simulation rig

✨ Same robot

✨ Same ROS 2 nodes

✨ Same commands

✨ Gazebo and Isaac Sim running side by side

✨ Behaviour compared using real numerical data

We begin with a migration that fails. The robot drives almost correctly but rotates only half as far as expected. We test friction, caster behaviour, joints, and PhysX settings until the real cause is isolated.

What you will investigate

✨ Geometry, mass, and inertia

✨ Joints and drive configuration

✨ Contact and friction

✨ Cameras and sensors

✨ ROS 2 interfaces

✨ Controllers and Nav2

✨ Full Gazebo to Isaac Sim behaviour

You will also learn to read divergence plots and recognize patterns such as rate errors, limits, frame offsets, geometry problems, and discrete events.

The course uses

✨ROS 2 Jazzy

✨Gazebo Harmonic

✨NVIDIA Isaac Sim 6.0.1

Every important result comes from reproducible scripts included in the repository. Failures are deliberately created, measured, debugged, and fixed on camera, including cases where the first hypothesis turns out to be wrong.

The goal is not just to import your robot into Isaac Sim. It is to make surethe migrated robot actually behaves like the original.

Who this course is for
⭐ ROS 2 roboticists with a robot already working in Gazebo who need it in Isaac Sim, and need proof it still behaves the same rather than a hope that it does.
⭐ Simulation and robotics engineers told to move to Omniverse for photoreal sensors, GPU-scale physics or synthetic data, who own the risk if a tuned stack breaks.
⭐ Anyone who has tried the URDF importer, got a robot that looked right and behaved differently, and could not find where the difference came from.