The navigation graph that houses the waypoints, are generated using the traffic_editor and building_map_tools packages in https://github.com/osrf/traffic_editor. This allows users to draw up their environment starting from a 2D floor plan, populating it with walls and models to best represent the real physical environment. With a close-enough representation of the real environment, lanes and waypoints can be placed in areas where the robots can navigate, while being mindful that lanes don't intersect with any existing furniture, for example. This allows users to generate a perfectly viable global navigation graph without a cost map.
The slotcar plugin allows us to simulate the motion and behavior of robots without worrying about the navigation stacks that might be running under the hood of different vendor robots. This will also allow us to run large scale simulations without being bogged down by large number of sensors and multiple navigation stacks running on that single machine.
If users are keen on running their own navigation stacks on simulated robots, I believe navigation2 is a good option to start with, I have had good experience with it on Turtlebot3s.