tf2 lookup_transform target_frame does not exist

edit

Try this. The commented code is the first attempt that did not work.

              import asyncio
...
                tf_future = self.tf_buffer.wait_for_transform_async(
                    target_frame=pose_stamped.header.frame_id,
                    source_frame=reference_frame,
                    time=rclpy.time.Time()
                )

                rclpy.spin_until_future_complete(self, tf_future)
                print("Got it!")

#                world_to_reference_transform = self.tf_buffer.lookup_transform(
#                    pose_stamped.header.frame_id,
#                    reference_frame,
#                    rclpy.time.Time(),
#                    timeout=rclpy.duration.Duration(seconds=5.0))

                world_to_reference_transform = asyncio.run(self.tf_buffer.lookup_transform_async(
                    pose_stamped.header.frame_id,
                    reference_frame,
                    rclpy.time.Time()
                ))

I read somewhere that this might be because the tf_buffer needs time to "fill up", but I haven't found a way to give it that time. The only way I could get things to work is by creating a timer and setting the get_dist() method as a callback. However, since I'm just looking up a static transform once, that approach doesn't seem right.

That's correct.

You create the TransformListener and then immediately query it with zero delay.

    self.tf_listener = TransformListener(self.tf_buffer, self)
    self.dist_gps = self.get_dist()

As such it will likely have received zero messages from any broadcaster not to mention a specific one that you're looking for.

A standard approach for this sort of thing would be to try to look up the trasnform and if it fails, wait a little bit and try again. Once you have confirmed receipt your startup process has completed.

However at the higher level I would question if this is the right approach overall. You are explicitly baking in your assumption about the source of the specific coordinate frames to your algorithm. A more robust way to approach this is to simply query the TransformListener for the transform when you need it, and don't make an assumption about the source or structure of the tree nor the frames that you're querying about. You can then make the frames parameters too. I bet that you won't see a significant performance improvement from precaching this one transform.

Of course when you do this make sure to use a try block and catch an error at or near startup before the TransformListener has received enough messages.