tf_static frames not available in lookup_transform

edit

My answer may be broader than what OP is asking, but because the initial motivation was the same for me to start tackling (static_transform not looked up via Python api in ROS2 for me), I share here what I found. It's embarrassing to admit that simple mistake.

Simply, I just had to run lookup AFTER spin at least once.

The following code is verified to lookup with turtle_tf2_py tutorial along with online tutorial (docs.ros.org/en/galactic/Tutorials/Intermediate/Tf2). Worked with static_transform too.

Btw, I still haven't had "try2" approach working (the one @AndyZe suggested). I doubt there _may_ be some issues as I see rclpy#989? No idea.

#!/usr/bin/python3

import asyncio
import time
import yaml

import rclpy
import tf2_ros
from tf2_ros import TransformException

class TfFramesFinder(rclpy.node.Node):
    def __init__(self):
        super().__init__("tf2_frames_finder")
        self._tf_buffer = tf2_ros.buffer.Buffer()
        self._tf_listener = tf2_ros.transform_listener.TransformListener(self._tf_buffer, self, spin_thread = False)

    def try3(self, target_f="kts2_table_frame", source_f="world"):
        self.get_logger().info("Getting it3!")
        _MAX_ATTEMPTS = 15
        _attempts = 0
        _transform_res = None
        while _attempts < _MAX_ATTEMPTS:
        _time_snapshot = rclpy.time.Time()
            try:
                self.get_logger().info("Looking up {}th time".format(_attempts))

                _transform_res = self._tf_buffer.lookup_transform(
                    target_f,
                    source_f,
                    _time_snapshot)
                break
            except TransformException as ex:
                if _MAX_ATTEMPTS < _attempts:
                    raise RuntimeError(
                        f"Could not transform '{target_f}' to '{source_f}' at the time '{_time_snapshot}'.: {ex}")
                else:
                    self.get_logger().warn("Failed {} times".format(_attempts))
                    time.sleep(1)
            _attempts += 1
        self.get_logger().info("x: {}, y: {}".format(
            _transform_res.transform.translation.x, _transform_res.transform.translation.y))

    def try2(self, target_f="kts2_table_frame", source_f="world"):
        self.get_logger().info("Getting it!")
        tf_future = self._tf_buffer.wait_for_transform_async(
            target_frame=target_f,
            source_frame=source_f,
            time=rclpy.time.Time())
        self.get_logger().info("wait_for_transform_async returned. tf_future: {}".format(tf_future))
        rclpy.spin_until_future_complete(self, tf_future)
        self.get_logger().info("spin_until_future_complete returned.")
        world_to_reference_transform = asyncio.run(self._tf_buffer.lookup_transform_async(
            target_f, source_f, rclpy.time.Time()))

rclpy.init()
node = TfFramesFinder()
#frame_parent = "turtle2"
frame_parent = "world"
#frame_child = "turtle2"
frame_child = "static_3"
try:
    rclpy.spin_once(node)
    node.try3(
        target_f=frame_parent,
        source_f=frame_child)
except KeyboardInterrupt as e:
    print("k/b interrupted")

node.destroy_node()
rclpy.shutdown()

I found a workaround after about 8 hours of tears. 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()
                ))

The lookup shouldn't care about whether the frame is static or not for detecting it. The best way to think of "frame not found" is that the buffer hasn't heard about that frame yet.

Without seeing anything more specific my first guess is that you are not giving your buffer enough time to initialize and connect on all the topics and for the static transform publisher to publish the static frame information. This will cause a race condition which will give you intermittent behavior.

The most common fix for this is to catch the exception and try again on the next cycle, or explicitly give time to initialize the transform buffer. But making your algorithm robust to the occasional transform data failure is strongly recommended if possible. This is a common race condition on startup of a distributed system.