Loop clousure in nav2d
Hello,
I'm wondering is there any loop closure detection in nav2d? if so, how can I use it.
Thank you Iman
add a comment
Thank you so much for your reply. One of the biggest problems I have is, when we talk about SLAM it means doing the mapping and localize it in the same time but all the method I can see in ROS so far is not close to SLAM. For example, first with gmapping we have to make the map and after that, we need to load that map with map server and with amcl localize the robot inside this map. It's not like simultaneous. I read in some post the nav2d can do the mapping and localization at the same time. Is that true? If not, what are the package I have to use for doing real SLAM and I don't need to save the map first and localize it offline? Thank you Iman
Yes, OpenKarto is SLAM, as is gmapping and others as well. While the map is build, the robot is constantly localized. You see this because the mapper (e.g. from nav2d_karto) publishes the pose of the robot in map coordinates. (Well, it actually publishes the difference between odometry and map pose, but this is just an implementation detail). While the mapper is running, you don't need amcl. But when you stop the mapper, because your map is finished, then you need amcl to do the localization, which was done by the SLAM before.
The difference between SLAM and a standalone mapping and standalone localization package can be thought of like this
SLAM makes a map and localizes the robot in that map as a byproduct of how the mapmaking takes place. In order to create the map, an on-line optimization must take place with some idea of where the robot is in order to create that map (at least in graph slam). Since the on-line mapmaking procedure needs to have the position of the robot at various timesteps in order to create the map, localization (to some degree) is also achieved.
Standalone mapping requires the robot to be localized. Think about it - if the robot doesnt know where it is but it sees a bunch of sensor information, how can we place that sensor information into some kind of map structure without knowing any of the motion constraints between sensor measurements?
Standalone localization ...
