Fixed Frame [map] does not exist - is LaMa localization node supposed to publish /map --> /odom ?

[WilliamWoo45]

Hi! Firstly, thanks for providing such great SLAM / Localization work.

I was trying to implement the localization function of LaMa ROS based on my own occupancy grid map. I launched the command lines in different windows in the following order:

1. rosrun iris_lama_ros loc2d_ros scan:=base_scan
2. rosrun map_server map_server /opt/ros/kinetic/share/husky_navigation/maps/mymap.yaml
3. rviz
4. rosbag play --clock /home/williamwoo/Z_husky_S2.1/S2.1_new/husky2-pose-_2020-06-12-s2.1-way11-mrpt.bag

The map is successfully loaded, where in the command line saying that:

[ INFO] [1606198756.368682224]: Requesting the map...
[ INFO] [1606198761.232647620, 1592210365.736253594]: Localization parameters: d_thresh: 0.01, a_thresh: 0.20, l2_max: 0.50
[ INFO] [1606198761.454955949, 1592210365.938106792]: 2D Localization node up and running

I suppose the localization node is successfully running. However, there is an error saying that Fixed Frame [map] does not exist (as shown below).

The tf is:

The rosbag info is:

So I'm wondering how could I solve this problem? I suppose what the LaMa ROS localization node does is to publish /map --> /odom, right? The map that I used is mymap.yaml and mymap.pgm; the rosbag that I used is here.

Thanks and all the best!

Best Regards, William

[eupedrosa]

Hi, Thank you for using LaMa.

I think your problem is here: rosrun iris_lama_ros loc2d_ros scan:=base_scan. Your bag has a laser scan in the /scan topic, so you should use that topic. The /map frame does not exist because the localization never received a laser scan.

Please try running with rosrun iris_lama_ros loc2d_ros scan:=/scan and let me know if it works.

[WilliamWoo45]

Hi Eurico,

Thanks a lot for your speedy response. It works! I did not expect the problem could come from there :)

BTW, I did the global localization tests and found that, without specifying the correct initial pose, the base_link and odom frames would jump to the correct place for 1-2 seconds at the beginning of some of my rosbags. Then they would jump back to somewhere near the origin (0,0), and the localization failed (see examples below). Example 1:

Example 2:

But at some starting place near the origin of the map (0,0), LaMa Localization managed to do the global localization after the robot moving for around 3-5m (see the screenshots below):

So, I'm wondering does the current localization node actually support Global Localization (i.e. localization without specifying initial pose) to some extent?

Cheers!

Best Regards, William

[eupedrosa]

The current localization node does not have support for Global Localization. So you must provide an initial pose, or you can use rviz for that.

However, there is a global localization mechanism already implement in the base code of the localization algorithm. It does not requires an initial pose. It may find the correct pose, but is is not guarenteed. There is a hard-coded threshold that may not be suitable for all applications.

I may implement the global localization in the ROS node. It is not a dificult task. Stay tuned.

[WilliamWoo45]

Hi Eurico,

Noted with thanks. I'll stay tuned and I'm very looking forward to your updates on global localization in the ROS node, since it is quite necessary when the robot is lost track of or when the robot is kidnapped in the practical applications.

Thanks again and take care!

Best Regards, William

[eupedrosa]

@WilliamWoo45, you can now try a global localization method for the localization node 242c3fca21da5bf4a6cbb089b0e8f41d03767004.

Please notice that the ~gloc_thresh parameter may need to be adjusted to your application.

[WilliamWoo45]

Hi Eurico,

Thanks a lot for your speedy coding. Noted. I'll try it as soon as possible.

Best Regards, William

[WilliamWoo45]

Hi Eurico @eupedrosa ,

I've implemented the global localization approach. Its localization success rate is around 10% ~ 15% in the repetitive corridor environment (as shown previously) without specifying the initial pose. I believe it can achieve much better global localization performance in ordinary environments where distinct geometric features are sufficient, and with better adjustment/tunning of the ~gloc_thresh parameter.

Regarding the global localization ability, I think LaMa ROS performs pretty good and is around the same performance level with AMCL or MRPT_localization.

Thanks and all the best!

Best Regards, William