Design Question: Where Does The Business Logic Go?

edit

To know what granularity is good for your nodes, keep in mind that a ROS node first job is to make processes communicate together. Having multiple processes allow you to have several independent algorithms running in parallel. However, processes are heavy objects and you cannot have a lot of them at the same time. To run thousands of tasks in parallel alternative techniques such as threading or coroutine are more suitable.

On top of that there is an overhead when you are communicating data from one node to another. ROS relies on the network to do this. This induces additional computation compared to a single-process algorithm.

To reduce this overhead and reduce the number of concurrent processes running, ROS supports nodelets. Nodelets are a way to aggregate several ROS nodes into one single process. This allow you to run as many nodes as you want into one process while removing the communication overhead.

However, using this model still has a cost as implementing nodelets is more challenging than challenging a normal node. In this case, think about the development cost of your application, the additional feature that ROS may or may not provide you and try to avoid over-engineer your solution.

I've been considering writing my own tutorials because the ones on the wiki are sorely lacking. However, you can learn a lot by looking at and expanding the code given in the tutorials.

The Gazebo simulator is the way to go if you want to work with a virtual robot in ROS. It allows you to build a robot with virtual sensors and program it to respond to those sensors in the same way it would in the real world. It's a very strong tool that the DARPA robotics challenge teams use often. Notice that the Stage simulator (stage ros) is a 2D version that is much easier to use. However, since Gazebo is much more common and well-documented, I'd suggest starting there.

If you've learned the fundamentals of ROS programming (i.e., you know what the code in the tutorials does and can teleoperate a turtle by Fahad khan), check out the gazebosim tutorials; they've recently been modified and should be quite nice now. You may either follow the tutorials in sequence or go in the following order:

1. Figure out how to incorporate a simple wheeled robot into the gazebo. Write some code to respond to a joystick or keypress by sending commands to the virtual robot. In Gazebo, teleoperate the robot.
2. Attach a laser scanning sensor or a bumper to your virtual robot's world, as well as some obstacles. As you teleoperate the sensor around, you can see the output of the sensor published in ROS.
3. Now that you've created the structure, it's time to create the robot's brains. There's a lot of navigation work already done for a robot with a laser scanner that you can more-or-less plug and play from the move base navigation stack. Alternatively, you will build your own intelligence and habits. When you're not bound by costly hardware, the sky's the limit.