hi, I'm working for Summer Student Program of Navigation2 : reduce the number of nodes running in Nav2 (tickets in Nav2), due to Node is overhead in ROS2.

When i reduce amcl_node nodes, i notice that the constructor of TransformListener has a argument spin_thread with default value true, which create a internal node to subscribe tf topic and a dedicated thread to spin this node.

i think the best way is to use new callback group and executor with dedicated thread so that we could avoid creating internal node which is overhead.

• If spin_thread, then create new callback group, and use executor to spin callback group instead of node.
• If not spin_thread, then its on the application to spin (no change)

After implementing this feature, we can use TransformListener in Node (for example AmclNode) like this to avoid creating internal Node.

tf_buffer_ = std::make_shared<tf2_ros::Buffer>(get_clock());
tf_listener_ = std::make_shared<tf2_ros::TransformListener>(
   *tf_buffer_, shared_from_this(), true);

any discussion is welcome, and i will make a PR if this feature is good.