ROS PMC Minutes for September 1st, 2026

The ROS PMC meeting for this week was on Tuesday, September 1, 2026. The notes from the meeting are available at ROS PMC weekly meeting agenda.

ROS PMC Business

• Election ballot is out. If you are a voting member, please check your inbox and vote. Seven votes were in as of the meeting, and a two-thirds majority of all PMC Constituents is required.
  • The ballot covers promoting @skye.galaxy, @jmachowinski, and @andrew_symington to PMC member, and moving @nuclearsandwich and @audrow to emeritus status.
  • A second ballot for the proposed committers will follow, once repository scope has been attached to each candidate. @mjcarroll needs to confirm with a couple of the candidates which repositories they are interested in. Expected by the end of this week, with the aim of completing before ROSCon.
• REP 158: a schema for simulation assets from USD files. The authors would like this across the line before ROSCon and plan to discuss it there.
  • @mjcarroll is assembling the REP approval packet, covering the REP itself along with links to the reference implementation and the Discourse discussion. Input and reviews are welcome, particularly from anyone who has not looked at it yet.
  • The Physical AI SIG and the Gazebo PMC have both been asked to review, and Gazebo has raised no objection.
  • It sits under ROS rather than Gazebo because it is a broader simulation standard than Gazebo alone. Gazebo is unlikely to adopt USD immediately, but O3DE and Isaac both make heavy use of the format.
  • The REP is well developed and has had input from the relevant parties, so this is largely process from here. If you have an objection, now is the time to raise it.
• Deferred: forming a Bazel Working Group, led by @andrew_symington, until he is a full PMC member.
• Informal TGC meeting this week, attended by @ahcorde, @cottsay, and @Katherine_Scott. Most of the discussion concerned AI scraping and AI policy, specifically which policy the OSRA should adopt or reuse and how to handle careless use. Nothing concrete came out of it yet.
  • @ahcorde is happy to relay input to the TGC if you have views on which policy direction makes sense.
  • Debian recently concluded its own vote on this. @claraberendsen tracked it and reports they landed somewhere similar to our existing policy.
  • On our own policy, there was some discussion of whether AI attribution belongs in the pull request description rather than on commit sign-off lines. @clalancette noted that knowing what was generated, and by which model, remains genuinely useful information.

Regular Business

Buildfarm update

The buildfarm builds and tests ROS packages across all supported distributions and platforms. This section covers new issues, current priorities, and infrastructure changes that affect maintainers.

• New greenness report. Greenness is generally down this month other than Lyrical, though the headline number counts only fully green builds and not yellow ones. The expanded report behind the month number gives the fuller picture.
  • Rolling has some green this month, in the low double digits. Not good in absolute terms, but the trend is in the right direction.
  • Zenoh is green again, after sustained effort from @yadunund and @ahcorde.
• Issues
  • zenoh_cpp_vendor Cargo.lock and OPAQUE_TYPES_BUILD_DIR warnings, Windows only. A warning rather than a regression. @mjcarroll has taken it and will try to reproduce locally.
  • Windows 11 ros2_multicast test_api regressions on release jobs, which had appeared in 2025 and have started again. Presents as a buffer space fault affecting a large number of multicast tests.
    • Working hypothesis is that SIGTERM-based teardown on Windows leaves resources behind and we eventually exhaust multicast sockets on the host.
    • @claraberendsen hit the same wall on Windows during Lyrical work. Because Windows uses process isolation rather than full containers, sequential runs share the underlying networking on the host, so incomplete cleanup accumulates. Reviewing the agent’s build log showed it had run nightly repeated jobs for at least five days in a row, so this may be a combination of repeated runs on one agent and cleanup not doing enough.
    • @MiguelCompany noted that exiting cleanly rather than terminating would close sockets, and ruled out parallelism as a cause: nightlies run packages in parallel, but these tests are all within a single package and therefore sequential.
    • @clalancette pointed out we already do some cleanup, killing stray colcon processes and similar, and that it could be made more extensive, though a run may also be conflicting with itself rather than inheriting a bad state.
    • @mjcarroll will look at extending the CI cleanup pass and whether there is an internal fix.
  • Warning on ignoring the return value of rmw_ret_t, affecting all Linux builds including RHEL and aarch64. Traced to a test merged into an RMW implementation two days ago; clang is stricter about this. @miguelgonrod will open the issue and ping the author, and it should be an easy fix.
• Priorities
  • PTAL Buildfarm Issues Dashboard (16 open issues).

Waffle note-taker

Waffle is a weekly meeting where maintainers triage incoming issues and pull requests that haven’t been otherwise acted upon. One person per week is assigned to be the “waffle note-taker” to both take notes in the meeting to keep the meeting as efficient as possible.

• Sep 3, 2026
  • @skye.galaxy, covering while @andrew_symington is on paternity leave.

The Baffle Maker board did not regenerate in time for last week’s meeting, so the group ended up triaging stale items. Merging the board into ros2/pmc_scripts is now the priority, so that it is not dependent on one person’s account and others can debug it when generation fails.

Rosdistro assignments

Rosdistro is a weekly assignment to monitor incoming issues and pull requests in GitHub - ros/rosdistro. Two people per week are assigned to be the rosdistro maintainers.

• Sep 1, 2026 - Sep 8, 2026
  • @clalancette
  • @sloretz
• Generally in good shape at around 11 open pull requests, which is about as good as it has looked.
• rosdistro#53263: an aggregated collection of example robot URDFs. This started as a naming and licensing question and broadened considerably.
  • @christophebedard flagged the name as too generic and suggested prepending the originating lab’s name. The contents turn out to be directories rather than individual packages, so this is a single package aggregating copies of others.
  • @clalancette raised the more fundamental question of whether we should take it at all, since it conflicts with the originals such as ur_description and human_description, and a copy will inevitably drift from upstream. The repository itself states it does not intend to substitute for the original repositories, which invites the question of what it is for. Rosdistro is a curated set more like a Linux distribution than an open-upload index.
  • @sloretz noted we have accepted generic names before, particularly around interface definitions, where maintainers commit to stewarding the name and accepting contributions broadly. That would be a reasonable condition here.
  • @yadunund and @christophebedard both objected to “data”, since these are descriptions, and to “example”, since the contents are named commercial robots rather than illustrative ones. It reads more like an awesome-list.
  • Alternatives raised: a ROS metapackage, or simply a documentation pull request listing the resources.
  • @emersonknapp made the broader point that we have no stated content guideline or quality bar, so we relitigate this case by case. We should either state a policy or accept that we do not have one.
  • Actions: @mjcarroll to draft rough content guidelines for rosdistro submissions. @clalancette to comment on the pull request as part of rosdistro duty this week.

Release Management

Release Management tracks the status of upcoming releases, patches, and syncs across the active ROS distributions.

• @christophebedard (Humble)
  • Last sync Aug 7, 2026. Due, and planned for next week once current rebuilds settle.
• @yadunund (Rolling)
  • Automation PRs in flight: ros_buildfarm#1151 and ros_buildfarm_config#396. @claraberendsen is reviewing and will deploy to the test farm.
  • The package percentage threshold has been set to 80%, which leaves roughly 200 packages that could be removed. That matches the largest number @yadunund has removed during a manual sync, so it is being used as the starting point.
• @marcogg (Jazzy)
  • Last sync Jun 18, 2026. @marcogg is working through the regressions in preparation for a sync.
• @cottsay (Kilted)
  • Last sync Aug 21, 2026 (471 packages, 0 regressions).
• @sloretz (Lyrical)
  • Freeze and sync this week.

Working Group Updates

• ROS Office Hours
  • Last meeting: Sep 4, 2026
  • Next meeting: Sep 18, 2026, the Friday before ROSCon, at 9:00 am Pacific.
  • Now on a recurring every-other-week invite. Hosting can rotate; it does not have to fall to @skye.galaxy every time. Everyone is welcome.
• Client Library Working Group (@alsora)
  • Next meeting: Sep 4, 2026 (agenda)
• Accelerated Memory Transport Working Group (@ahcorde)
  • Next meeting: Sep 2, 2026
• ROSGraph Working Group (@emersonknapp)
  • Next meeting: Sep 1, 2026 (today). NoDL documentation
  • End-to-end C++ node generation from NoDL is working for the first time. Still needs ironing out, but the full round-trip workflow can now be prototyped.
  • Tutorials are going up on the doc site.
  • The bloom side is sorted, with the release team updated and rosdistro entries consolidated, so the current version can go into the buildfarm and be incremented from there.
  • The graph monitor now collects parameters, giving a live view of all topics and parameters in a running system. Services and actions are still to come. Next step is tying the packages together so the graph monitor can use the NoDL tooling.
• SIG Physical AI (reporting by @yadunund)
  • The SIG will now report to the PMC regularly. Start with the OSRA blog post, which sets out how the group is positioning ROS in an ecosystem where people increasingly generate their own middleware and integrations.
  • The SIG is a collective of people working on robot learning and end-to-end policies on top of ROS, consuming output from the working groups here and producing standards and reference implementations.
  • Demos show the same controller driving both a LeRobot SO-101 arm and a UR robot, with the difference handled through configuration and launch parameters, illustrating a path from hobby hardware to industrial.
  • WG1 (interfaces, interoperability, transport): the zero-copy rosidl concept has reached proof of life at roughly 1 kHz and 1 ms for 4 MB images over Fast DDS shared memory, with additional delay sources under investigation. The group is also reviewing REP 158. A Hardware Configuration Description Format proposal, combining URDF, SDF, and hardware-specific parameters, has a Discourse post pending. OpenUSD interop is still collecting feedback. The 3D LiDAR message is stalled and the tactile proposal is gathering feedback.
  • WG2 (data recording and replay): no updates. Building on Rosetta, a community integration between rosbag2, ROS topics, and the LeRobot ecosystem, which uses Parquet.
  • WG3: focused on the ROSCon talk.
  • WG4 (applications): migrating demos from Kilted to Lyrical. The AIC cable-insertion demo is in a draft PR and will be reorganized into per-application subfolders. Next steps are integrating WG3’s inference pipeline, adopting the CUDA buffer with ONNX converters, and migrating to the new Rosetta structure.
  • Roadmap: a proposal is being put together for a reference application with a more complex embodiment, such as a dual arm on a mobile base or a humanoid, intended to line up with the next release as a flagship demonstration platform.
  • How to get involved: the meeting is on the OSRA calendar and there are public Zulip channels. Links are at the bottom of the blog post.

Agenda

• [← 返回资讯流