What's the recommended project structure for a basic ROS project that uses Gazebo for simulation? I'm reading through the ROS and Gazebo tutorials, as well as looking at some projects like Turtlebot, and I'm having trouble figuring out what the best practice is. In some ways, Turtlebot's project layout seems very complicated, even when compared to physically more complicated robots like the Nao, whose project layout is relatively more simple.

I'm trying create an initial project layout for a small quadruped robot that I plan to simulate in Gazebo. This tutorial recommends creating just two packages, MYROBOTdescription and MYROBOTgazebo in a single git repo. Is that all I need? Looking at Turtlebot, they've done something similar, but they've split MYROBOTgazebo off into a turtlebotsimulator meta-package in a separate git repo. I understand the abstraction because it lets them plug in different simulator backends other than Gazebo, but I don't understand why they manage the source in a separate repo. What's the benefit of keeping the *description and *gazebo projects in separate repos?

Turtlebot also defines a few other packages, which I've seen replicated elsewhere, such as *bringup, *capabilities, and *_apps. Are these typically best practice to create for your robot?