Does anyone know how do the link between Rviz of ROS and the physical robot model ?

edit

• This is usually done by creating a URDF (robot kinematic model) which defines the kinematics and joints of your robot, and loading it into the robot_description parameter.
• Your hardware driver then publishes the joint_states topic with the name and position of each joint mentioned in the URDF.
• The robot_state_publisher node reads this joint_states topic and the URDF, and publishes the transform tree (TF) in full 3D for your robot.
• You can then visualize this in Rviz with the Robot Model display.

The Using URDF with robot_state_publisher Tutorial gives a more in-depth walkthrough of all of the steps in this process.