Why is Nav-Stack
so much buggier with the addition of namespaces?
Hello, I have been working with two turtlebot3 burgers and have been successfully implementing my code and packages that I need. However, when I run the same code with the same parameters but I put just a single robot into its own namespace, it results in the general performance of the navstack being severely reduced. The code still works but I find it much more sensitive to things like running on a fresh roscore and robot bringup. The main issues I run into is delays with the movebase simple action client and amcl not tracking well. Has anyone seen this before? Does adding a namespace really increase the computational overhead that much?
Suggestion: change your question title to better reflect your question. "ROS" is a) very ambiguous (as it would include all available software across all available repositories) and b) your question text actuallya describes observations with a limited set of packages.
May be you can add some statistics? I have not heard of this behaviour before. Perhaps you can add frequency of control messages (simple
rostopic hz) from move_base before and after the namespace addition. Also statistics like CPU usage or RAM usage. Ensure the inputs for both tests are the same (same starting point, same environment, same costmap, etc) with fresh restarts of roscore inbetween both tests.Sounds good, I actually found a rather strange work around. If I launch the same node without a namespace and then bring up another in a namespace everything works perfectly fine. Both robots are happy etc. and navigate seamlessly. Before the fix though the local costmap often fails to load at all. Amcl pose however is being published at 7hz and the controller and planner frequencies are at 5hz, laser scans are coming in at 5hz.