I am working with the turtlebot simulator. I have subscribed to the topic /Odometry to read the message on it. I compute the difference between the location in this message and the ground truth location of the robot. I subscribe to the /gazebo/modelstates topic to obtain the ground truth location (the "mobilebase" component of the message).

This is the plot of the difference (2D euclidean distance) between the Odometry message and ground truth location: (the sample number increases linearly with time, so the x-axis is proportional to simulation time)

Are there parameters that can be tweaked to reduce this level of accuracy? The problem is that the odometry seems to be accurate within a few centimeters, which is unrealistic. This limits what we can study using the simulated model.