Add brief overview of pc-robot network configuration

[erelson]

@mmwise Is this what you had in mind?

[mikeferguson]

@erelson -- there is something a little more subtle going on here than is evident in your drawing and text, which are a bit misleading:

• The roscore, which runs on the robot, will actually use the hostname as the rosmaster IP, for instance a robot whose hostname is "fetch1", would be accessible at "http://fetch1:11311".
• When you start a new terminal session and source a setup.bash file, you end with various defaults -- one of those is that ROS_MASTER_URI is set to http://localhost:11311. As it so happens, on the robot, localhost:11311 will resolve properly and you can query the rosmaster.
• What happens after you query is where things typically get interesting:
  • Say that I have a node "A" which wants to subscribe to a topic "topicB" which is being published by a node "B" that is locally on the robot. If I am running "A" on the robot, then my ROS_MASTER_URI with localhost resolves, I get a response from the roscore, but that response tells me that "topicB" is located at some port on the robot, let's say "freight1:5411". Since "A" runs on the robot, freight1 resolves to localhost (since it is in the /etc/hosts) and everything works fine.
  • However, let's now say that node "A" is running on my desktop computer. I've pointed ROS_MASTER_URI to the IP Address of the robot (say "192.168.0.10:11311"), I get a hold of the roscore without issue, I get a response from it, but then my local computer has no idea how to connect to "freight1:5411", and so I get no data.
    • This second case is also why things like "rosservice list", "rostopic list" work fine, since all the data is carried back through the original connection made by my computer to "192.168.0.10:11311", but "rostopic echo topicA" fails to return anything because a new connection has to be made to "freight1:5411" for any topic data to be received.

The bottom line is: you really need to be able to get to the robot by name, using IP addresses in the ROS_MASTER_URI is generally a bad idea, you should use the robot hostname. If there is no DNS available to lookup the name in the environment, an /etc/hosts entry should be added that maps FetchX to it's IP address.

P.S. People inevitably ask "why not just map the roscore to use IP:11311"? Some people do this, and you will find numerous tutorials that tell you to do this, but this has a number of downsides:

• If the IP changes, the entire roscore and every ros process on the robot has to be restarted.
• If we use the IP, and the link drops temporarily, if our /etc/hosts has not mapped that IP to localhost, we may have packets get stuck in transit until the link returns. This can lead to a variety of failure cases that are very hard to diagnose.