TF Tree in blue stack

[Manouselis]

Hi everyone,

I am trying to implement a custom position controller. To do that, I need to transform my position error from the map frame to the base_link frame. When trying to do that I noticed that there is no TF Connectivity between map and base_link. By inspecting the TF tree, I also saw that map, odom and base_link are all disconnected from each other. I would normally expect a transformation similar to: map --> odom --> base_link. Why is that not the case?

frames_disconnected.pdf

[evan-palmer]

The easiest way to achieve this is to add a ROS bridge from Gazebo. This is the solution that I used when implementing the new demos for the UVMS controllers here.

The map -> odom link doesn't exist in this simulated system because it uses the absolute Gazebo state measurements. In practice, when you are using a USBL + DVL, those TFs are more important due to the low sampling rates of the sensors.

[Manouselis]

Hey @evan-palmer, this worked, thank you. I'm now wondering how this will work in the actual physical system. In the physical system we'll only have a local position estimate. We therefore won't have a transformation between the "map" and the "base_link_frd" frames. This transformation is needed for topics such as the /mavros/local_position/odom topic which is in "map" frame. Did I misunderstand something or did you tackle this problem somehow?

[evan-palmer]

On hardware, we obtain the odom -> base_link transform using our state estimator (robot_localization). We treat the odom frame as our psuedo-map frame. However, if you happen to receive a periodic GPS lock, you could likely define a true map -> odom transform.

With regards to MAVROS and ArduPilot, I previously spent time trying to integrate the ArduPilot EKF (exposed in ROS via the /mavros/local_position/odom topic); however, I ran into issues integrating additional sensor modalities and configuring it in a way that worked for my specific use-case. This encouraged my adoption of the robot_localization package.

[Manouselis]

Thanks for the quick reply @evan-palmer. Can you guide me to where exactly the robot_localization package was implemented in the blue repo (or in the docker image), as I can not find any of the topics I expect from robot localization such as odometry/filtered or nodes like the ekf_localization_node?

[evan-palmer]

I haven't included that implementation in this project because the configuration is tailored specifically to our own hardware deployment. To integrate it with your AUV, you can simply change the endpoint from which you receive odometry estimates (i.e., instead of receiving odometry from Gazebo, you receive it via odometry/filtered. An example demonstrating how to configure the EKF node is here.

[evan-palmer]

As a quick side comment, you may find the mavros_controllers package helpful as a reference for your position controller.