[TomCC7]

What is this feature?

Publish applied_joint_position_command JointState events after successful POSITION/SERVO_POSITION hardware writes. Events contain only accepted joint targets and the control snapshot timestamp; publication occurs outside the hardware lock.

Why do we need this?

Dataset actions need accepted command history rather than measured state or commands that hardware rejected. This is a prerequisite of #4343, stacked after #4390 and immediately before the integration PR.

How to Test

uv run pytest dimos/control/test_control.py dimos/control/test_per_robot_joint_states.py

97 control tests passed, including accepted/rejected writes and sparse target publication; commit hooks passed. Real hardware execution has not been verified.

Which issue(s) does this PR close?

None. This is a review slice of #4343, following the Manipulation SDK preview. It does not close or replace the integration PR.

Checklist

• I have read and approved the CLA.

AI assistance: Codex assisted with the split, validation, and description.

Inherited review fixes

Inherit the #4395 complete-dataset publication and inspection fixes, and the #4396 recorder-local port declarations. Core Module and Blueprint definitions are unchanged, and the collection stack audit script is removed. This layer retains its existing control behavior and PR base.

The final assembled stack passed 278 host tests with two native-recorder tests excluded by the default marker filter. The LeRobot dataprep and MCAP suite passed 28 tests using cached LeRobot 0.6.0 and datasets 4.7.0; this is not a full locked-environment run. Scoped mypy passed for six scoped production files, and changed-file pre-commit passed. No dependencies were installed, no hardware was operated, and CI has not been checked.

Inherited training-key conversion

Inherit #4395's fixed LeRobot feature naming and MCAP-to-dataset E2E regression. Profiles retain their source declarations; conversion produces standard training-facing keys and records the mapping for inspection. No rename configuration is introduced. The final stack passed 278 host tests and 28 conversion/MCAP tests in the documented cached environment.

Rebased stack integration

Inherit #4395/#4396 rebased onto main, the recorder lifecycle fix, timestamp-boundary regression, and the Python-generated profile-to-dataset E2E. This layer keeps its original changes and existing history.

Current validation: 280 host tests passed with two pre-existing native-recorder tests excluded by the host marker filter. Thirty LeRobot conversion/E2E tests passed, including both Python-generated profile cases. Python tests used cached LeRobot 0.6.0 and datasets 4.7.0, not a complete locked isolated environment. The updated Python E2E passed scoped mypy; pre-commit passed. No physical hardware or policy training was used. CI has not been polled.

[github-actions]

go2 replay realtime

Details

Benchmark suite	Current: 1d029a8	Previous: cdfaef0	Ratio
first frame wall	9.3 s	9.891 s	0.94
first frame cpu	18.938 s	21.456 s	0.88
run cpu	29.88 s	33.75 s	0.89
run cpu (user)	22.164 s	26.257 s	0.84
run cpu (system)	7.715 s	7.493 s	1.03
peak memory	2712.355 MB	2699.734 MB	1.00
peak threads	404 threads	394 threads	1.03
disk read	8.137 MB	20.871 MB	0.39
disk write	11.137 MB	11.098 MB	1.00
network (transport)	2713.989 MB	2722.56 MB	1.00
network (external rx)	0.114 MB	0.139 MB	0.82
network (external tx)	0.154 MB	0.342 MB	0.45

This comment was automatically generated by workflow using github-action-benchmark.

[codecov]

❌ 1 Tests Failed:

Tests completed	Failed	Passed	Skipped
6866	1	6865	59

View the top 1 failed test(s) by shortest run time

dimos.codebase_checks.test_blueprint_kwargs::test_blueprint_atom_kwargs_match_module_config[learning-collect-webxr-piper]

Stack Traces | 0.005s run time

blueprint_name = 'learning-collect-webxr-piper'
monkeypatch = <_pytest.monkeypatch.MonkeyPatch object at 0x170499f90>

    @pytest.mark.parametrize("blueprint_name", _blueprint_params())
    def test_blueprint_atom_kwargs_match_module_config(
        blueprint_name: str, monkeypatch: pytest.MonkeyPatch
    ) -> None:
        """Fail when blueprint kwargs cannot be consumed by their target module."""
        # The multi-robot blueprints read ROBOT_IPS at import time.
        monkeypatch.setattr(global_config, "robot_ips", "192.0.2.10,192.0.2.11")
        blueprint = _get_blueprint_or_skip(blueprint_name)
    
        violations: list[str] = []
        for atom in blueprint.blueprints:
            unknown_kwargs = sorted(set(atom.kwargs) - _allowed_kwarg_names(atom.module))
            if unknown_kwargs:
                violations.append(
                    f"{atom.module.__module__}.{atom.module.__name__}: unknown kwargs {unknown_kwargs}"
                )
    
        if violations:
            listing = "\n".join(f"  - {violation}" for violation in violations)
>           raise AssertionError(
                f"Blueprint {blueprint_name!r} passes unknown module kwargs:\n{listing}\n\n"
                "Blueprint kwargs are forwarded into the module constructor. For modules "
                "with an explicit `config` annotation, use fields from that config model; "
                "for legacy modules with direct constructor parameters, use the declared "
                "`__init__` keyword names."
            )
E           AssertionError: Blueprint 'learning-collect-webxr-piper' passes unknown module kwargs:
E             - dimos.imitation.collection.recorder.CollectionRecorder: unknown kwargs ['input_types', 'recording_schema']
E           
E           Blueprint kwargs are forwarded into the module constructor. For modules with an explicit `config` annotation, use fields from that config model; for legacy modules with direct constructor parameters, use the declared `__init__` keyword names.

atom       = BlueprintAtom(kwargs={'enable_pointcloud': False}, module=<class 'dimos.hardware.sensors.camera.realsense.camera.RealS...me='tf', type=<class 'dimos.msgs.tf2_msgs.TFMessage.TFMessage'>, direction='out')), module_refs=(), instance_name=None)
blueprint  = Blueprint(blueprints=(BlueprintAtom(kwargs={'input_types': {'color_image': <class 'dimos.msgs.sensor_msgs.Image.Image'...'left_gripper_command'): 'left_gripper_command'}), requirement_checks=(), configurator_checks=(), hosted_placements=())
blueprint_name = 'learning-collect-webxr-piper'
listing    = "  - dimos.imitation.collection.recorder.CollectionRecorder: unknown kwargs ['input_types', 'recording_schema']"
monkeypatch = <_pytest.monkeypatch.MonkeyPatch object at 0x170499f90>
unknown_kwargs = []
violations = ["dimos.imitation.collection.recorder.CollectionRecorder: unknown kwargs ['input_types', 'recording_schema']"]

dimos/codebase_checks/test_blueprint_kwargs.py:97: AssertionError

To view more test analytics, go to the Test Analytics Dashboard
_{📋 Got 3 mins? Take this short survey to help us improve Test Analytics.}

[github-actions]

go2 replay realtime (arm64)

Details

Benchmark suite	Current: 55e5f9e	Previous: d793e80	Ratio
peak memory	1740.984 MB	1716.062 MB	1.01
peak threads	374 threads	367 threads	1.02
network (transport)	2730.918 MB	2724.058 MB	1.00
disk write	1.684 MB	1.738 MB	0.97
instructions	137.561 G	137.102 G	1.00

This comment was automatically generated by workflow using github-action-benchmark.

[TomCC7]

I think we already have support for this in main?