examples(teleop_ros2): package under isaacteleop_examples

[jiwenc-nv]

Description

Part of #985. Stacked on #1038.

Fifteen modules imported their siblings as top-level modules — constants, messages, geometry, assets, session_config and the rest — which resolved only because the invoked script's own directory landed on sys.path. That broke under python -m, broke when copied into another project, and claimed an unusually generic set of names. This moves the tree to the layout in examples/README.md.

Two workarounds propping this up are deleted, which is the evidence the packaging is real rather than relocated:

• pythonpath = ["."] in the pyproject
• the example source directory on the ctest PYTHONPATH

The tests reach the in-tree source through a conftest.py instead, mirroring mujoco_xr's, so a bare pytest in that directory works for the first time. Six test import lines change; no assertion does.

The node keeps its filename and gains a __main__.py, so the container entry point becomes python -m isaacteleop_examples.teleop_ros2 and still forwards --ros-args. The install tree drops a level, so the Dockerfile WORKDIR follows, as do the three invocations in build-ubuntu.yml.

Type of change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to change)
• Documentation update

Testing

x86_64 / Ubuntu, Python 3.12. Wheel ships 15 modules under isaacteleop_examples/teleop_ros2/ with no namespace __init__.py. Editable-installed into a clean venv, the ROS-free modules (constants, tensor_group_helpers) import with the CWD outside the repo; import constants fails, so the flat names are gone rather than relocated. The new conftest.py was checked to resolve to examples/teleop_ros2/python with the package present and no namespace __init__.py. All test files compile. pre-commit clean.

Not run: ctest -L teleop_ros2 — no ROS 2 on this host, so geometry_msgs and rclpy are unavailable. That suite and the container entry point are what CI should be trusted for here; please treat the CI run as the gate rather than my local checks.

Checklist

• I have read and understood the contribution guidelines
• I have run the linter and formatter with SKIP=check-copyright-year pre-commit run --all-files
• I have made corresponding changes to the documentation
• I have added tests that prove my fix/feature works (or explained why not)
• I have signed off all my commits (git commit -s) per the DCO

_{No new tests: this is a move. The existing suite is the check — it exercises every moved module and passes with only its import lines changed.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: d58e75e3-b5cc-4acc-a175-a979325602f1

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

_{Comment @coderabbitai help to get the list of available commands.}

[ivany-nv]

A heads-up about this PR's base rather than its own content: the branch is still
built on the pre-rebase copy of the mcap_record_replay packaging commit, so it
carries versions of files that #1038 has since fixed.

This head (774a9f0f4) sits on 091b91cb2 examples(mcap_record_replay): package under isaacteleop_examples and does not contain #1038's current head
887db467. The same holds all the way up the stack — #1047 also resolves to
091b91cb2.

So the three commits #1038 picked up on 2026-09-08 are absent here:

• 681bc222b — bind viser viewers to every interface by default
• 0981f53ff — filter replay auto-discovery by recording type
• 887db467f — ignore /recordings/ written from the repo root

Checkable on the two heads:

$ git show 774a9f0f4:examples/mcap_record_replay/python/isaacteleop_examples/mcap_record_replay/replay_full_body.py | sed -n 66p
    candidates = list(recordings.glob("*.mcap"))

$ git show 887db467f:examples/mcap_record_replay/python/isaacteleop_examples/mcap_record_replay/replay_full_body.py | sed -n 66p
    candidates = list(recordings.glob("full_body_*.mcap")) or list(

$ git show 774a9f0f4:.gitignore | grep -c '^/recordings/'
0

Two consequences:

1. Merging up the stack without a rebase would revert those fixes — the
   type-prefixed auto-discovery in particular, which is the one that stops
   replay_full_body from picking a controllers_*.mcap.
2. Because the merge-base is old, this PR's diff currently renders as 50 files
   / +310 −137, most of which is the stale mcap_record_replay content rather
   than the teleop_ros2 change actually under review.

A rebase of #1039..#1047 onto the current #1038 head would clear both. I held
off reviewing the teleop_ros2 change itself for now, since the diff as shown
is not what would land — happy to pick it up once the base is current.

[ivany-nv]

Verdict: a rebase onto #1038's current head looks unavoidable. #1038 was rebased after this stack branched, and both merge orders now conflict on the same three files — including on the very fix #1038 added, where a careless resolution would quietly undo it.

Concrete evidence for the baseline point raised earlier. The stack's 091b91cb2 is not an ancestor of #1038's current head 887db467. It was replaced by f7b37f939, followed by three more commits — one of which, 0981f53ff "filter replay auto-discovery by recording type", is the fix for the defect reported in the #1038 review.

Both merge orders conflict, on the same three files (replay_controller.py, replay_full_body.py, replay_hand.py):

order	result
#1038 then stack	#1038 clean, stack 3 conflicts
stack then #1038	stack clean, #1038 3 conflicts

So sequencing alone will not avoid it. The part worth flagging: the conflict sits on the fix itself — a merged file holds both glob("full_body_*.mcap") or glob("*.mcap") and the older bare glob("*.mcap"), so whoever resolves it could revert the fix without noticing.