ROS 2 Gazebo Simulation Guide 2026

Simulation is where robots are built before hardware exists. This guide covers modern Gazebo (gz-sim Harmonic) with ROS 2: authoring worlds, simulating sensors, tuning physics, bridging topics, and closing the sim-to-real gap.

1. Installing Gazebo Harmonic + ROS 2

Modern Gazebo (formerly "Ignition") pairs with ROS 2 via ros_gz:

# Gazebo Harmonic (gz-sim 8) with ROS 2 Humble/Jazzy
sudo apt-get install -y   ros-humble-ros-gz   ros-humble-ros-gz-sim   ros-humble-ros-gz-bridge   ros-humble-ros-gz-image

# Launch an empty world
gz sim empty.sdf

# Verify the ROS <-> gz bridge tooling
ros2 pkg list | grep ros_gz

2. Authoring an SDF World

Worlds are SDF. Define physics, lighting, and models:

<?xml version="1.0" ?>
<sdf version="1.10">
  <world name="warehouse">
    <physics name="1ms" type="ignored">
      <max_step_size>0.001</max_step_size>
      <real_time_factor>1.0</real_time_factor>
    </physics>

    <plugin filename="gz-sim-physics-system"
            name="gz::sim::systems::Physics"/>
    <plugin filename="gz-sim-sensors-system"
            name="gz::sim::systems::Sensors">
      <render_engine>ogre2</render_engine>
    </plugin>

    <light type="directional" name="sun">
      <direction>-0.5 0.1 -0.9</direction>
    </light>

    <include>
      <uri>model://warehouse_shelves</uri>
    </include>
  </world>
</sdf>

3. Simulating Sensors

Attach a 3D LiDAR and a depth camera to the robot in SDF:

<sensor name="lidar" type="gpu_lidar">
  <update_rate>10</update_rate>
  <topic>scan</topic>
  <lidar>
    <scan>
      <horizontal><samples>360</samples>
        <min_angle>-3.14</min_angle><max_angle>3.14</max_angle>
      </horizontal>
      <vertical><samples>16</samples>
        <min_angle>-0.26</min_angle><max_angle>0.26</max_angle>
      </vertical>
    </scan>
    <range><min>0.1</min><max>30.0</max></range>
  </lidar>
</sensor>

<sensor name="depth" type="depth_camera">
  <update_rate>30</update_rate>
  <topic>camera/depth</topic>
  <camera><image><width>640</width><height>480</height></image></camera>
</sensor>

4. Bridging gz Topics to ROS 2

ros_gz_bridge maps Gazebo transport topics to ROS 2 message types:

# Bridge specific topics (gz_type <-> ros_type)
ros2 run ros_gz_bridge parameter_bridge   /scan@sensor_msgs/msg/LaserScan[gz.msgs.LaserScan   /camera/depth@sensor_msgs/msg/Image[gz.msgs.Image   /cmd_vel@geometry_msgs/msg/Twist]gz.msgs.Twist   /model/robot/odometry@nav_msgs/msg/Odometry[gz.msgs.Odometry

# '[' = gz -> ros, ']' = ros -> gz, '@' separates names from types.
# For many topics, use a YAML bridge config instead.

5. Spawning Robots at Runtime

from launch import LaunchDescription
from launch_ros.actions import Node
from ros_gz_sim.actions import GzServer

def generate_launch_description():
    return LaunchDescription([
        GzServer(world_sdf_file='warehouse.sdf'),
        Node(package='ros_gz_sim', executable='create',
             arguments=['-name', 'robot',
                        '-file', 'robot.sdf',
                        '-x', '0', '-y', '0', '-z', '0.2'],
             output='screen'),
    ])

6. Physics Tuning for Fidelity

• Step size: 1ms (1000Hz) for contact-rich manipulation; larger for mobile bases.
• Solver: raise iterations for stable grasps and stacking.
• Friction: match real surface coefficients — the #1 sim-to-real gap for wheels/feet.
• Inertia: correct mass/inertia tensors or the robot behaves nothing like reality.
• Engine: DART default; Bullet/TPE for speed vs accuracy trade-offs.

7. Sim-to-Real: Domain Randomization

Randomize sim parameters so a policy trained in sim survives reality:

import random

def randomize_episode(world):
    world.set_friction(random.uniform(0.6, 1.2))
    world.set_gravity_z(random.uniform(-9.7, -9.9))
    world.set_light_intensity(random.uniform(0.5, 1.5))
    world.add_sensor_noise('lidar', stddev=random.uniform(0.0, 0.03))
    world.set_actuator_latency(random.uniform(0.0, 0.02))
    # Training across this distribution makes the policy robust to
    # the unknown true parameters of the real robot.

8. Headless Sim for CI & RL

# Headless server (no GUI) for cloud training / CI pipelines
gz sim -s -r --headless-rendering warehouse.sdf

# Run many parallel instances with distinct partitions
GZ_PARTITION=env0 gz sim -s -r world.sdf &
GZ_PARTITION=env1 gz sim -s -r world.sdf &
# Each is isolated — collect experience in parallel for RL.

9. Validating Against Reality

• Log comparison: replay a real rosbag's commands in sim, compare trajectories.
• Sensor realism: add noise/dropout models matching your real sensors.
• Latency: inject the real control/comms delay into sim.
• Contact: validate grasps/footsteps against real force data.

10. When to Use Gazebo vs Isaac Sim

• Gazebo: open-source, great for navigation, multi-robot, and CI; lighter compute.
• Isaac Sim: photorealistic perception, GPU RL at scale, NVIDIA hardware.
• MuJoCo: fastest contact dynamics for locomotion/manipulation RL.
• Many teams prototype in Gazebo, then train perception/RL in Isaac.

Key Takeaways

Modern Gazebo (gz-sim Harmonic) is the open-source backbone of ROS 2 simulation: author SDF worlds, simulate LiDAR and depth sensors, and bridge to ROS 2 with ros_gz_bridge. Tune physics — especially friction and inertia — to shrink the sim-to-real gap, and use domain randomization plus headless parallel sim for robust RL. Validate against real rosbag logs before trusting sim, and reach for Isaac Sim or MuJoCo when you need photorealism or fast contact dynamics.