Simulation Description Format (SDF, also called SDFormat) is Gazebo’s native XML format for describing simulated worlds and individual objects. It plays a role similar to URDF: both describe robots, but SDF also describes the surrounding world. In ROS–Gazebo workflows, a common convention is to keep the robot in URDF and use SDF for the world and other simulated objects. section 5.3 section 5.6
An SDF model groups physical parts as links and connects them with joints. The simulated physics engine treats links and environmental objects as rigid bodies; joints constrain their relative motion. Meshes provide geometry for collision detection, while textures improve visual appearance. section 5.3
<sdf version="1.9">
<model name="MODEL_NAME">
<link>...</link>
<joint>...</joint>
...
</model>
</sdf>
A minimal model starts with an SDF version, then names a model containing links and joints. This is the basic structure for a simulated object. section 5.3
A world file can include models at named poses, making SDF useful for assembling a scene rather than only defining one robot. Sensors and actuators extend the environment–sensor–controller loop: sensor tags attach to links, while ros2_control tags associate control interfaces with joints. Gazebo plugins implement the resulting simulated devices and behaviors. section 5.3
Available plugin-backed simulation components include:
- Touch, force-torque, RGB, RGBD, thermal, lidar, IMU, and GNSS sensors. section 5.3
- Joint control by torque, velocity, or position. section 5.3
- Multirotor, wheeled, tracked-vehicle, and conveyor-belt control models. section 5.3
<sdf version="1.9">
<world name="WORLD_NAME">
...
<include>
<uri>file://PATH/TO/MODEL</uri>
<name>NAME_OF_INCLUDED_MODEL_IN_WORLD</name>
<pose>0 0 0 0 0 0</pose>
</include>
</world>
</sdf>
A world can include a model from a file URI, assign it a name in the world, and place it with a six-value pose. section 5.3
Camera-equipped robot simulation
A camera sensor can be associated with a robot link and configured with field of view, image size, clipping range, update rate, visualization, and a topic. A camera on the rrl_bot end-effector then produces an image that can be visualized in RQt, connecting a simulated physical pose to a ROS-facing sensor stream. section 5.3
SDF supports practical simulation work, but model fidelity still matters. Rigid-body and point-contact assumptions in physics engines, along with rendering and visual attributes, contribute to the sim-to-real gap. Gazebo can also combine models through co-simulation, including Functional Mock-up Units for components such as electric motors. section 5.6 section 5.3
