refactor(imitation): isolate LeRobot policy runtime by TomCC7 · Pull Request #3315 · dimensionalOS/dimos

TomCC7 · 2026-07-31T20:41:34Z

This is layer 2 of the OpenYAM learning slice in the original stack.

Problem

LeRobot requires an environment that conflicts with the main DimOS perception and development dependency set. The policy module still needs normal typed streams and RPC control from the host stack.

Solution

Define an RPC-only LeRobotPolicyModule host contract using IsolatedPythonModule.
Keep LeRobot, Transformers, Torch, runtime tests, and their lockfile in a sibling Python 3.12 project.
Preserve lazy checkpoint loading, live observation validation, ordered joint targets, concurrency control, status reporting, and clean shutdown.
Keep LeRobot dependencies out of the root extras and root lockfile.

Learning slice

feat: add isolated Python module runtime #3478 — external Python native-module runtime
fix(openyam): align learning stack with canonical model #3853 — OpenYAM learning baseline
refactor(imitation): isolate LeRobot policy runtime #3315 — isolated LeRobot policy runtime
feat(imitation): add native OpenYAM collection #3854 — native OpenYAM collection
feat(imitation): add controlled OpenYAM policy rollout #3855 — controlled OpenYAM policy rollout
feat(a1z): a1z learning workflow #3318 — A1Z learning workflow

Validation: 21 host-side tests and 14 isolated-runtime tests passed.

Checklist

I have read and approved the CLA.

codecov · 2026-07-31T20:48:57Z

❌ 1 Tests Failed:

Tests completed	Failed	Passed	Skipped
4687	1	4686	180

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

dimos.robot.manipulators.openyam.test_openyam::test_make_openyam_model_config_uses_canonical_arm_joints

Stack Traces | 3.23s run time

def test_make_openyam_model_config_uses_canonical_arm_joints() -> None:
        config = make_openyam_model_config()
    
>       assert OPENYAM_MODEL_PATH.parts[-2:] == ("i2rt", "yam.urdf")

config     = <[RuntimeError("Failed to pull LFS file .../dimos/data/.lfs/yam_description.tar.gz after 3 attempt...tar.gz', '--exclude=']' returned non-zero exit status 1.") raised in repr()] RobotModelConfig object at 0xff1d30df7b60>

.../manipulators/openyam/test_openyam.py:63: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
dimos/utils/data.py:370: in __getattribute__
    resolved = object.__getattribute__(self, "_ensure_downloaded")()
        name       = 'parts'
        self       = <[RuntimeError("Failed to pull LFS file .../dimos/data/.lfs/yam_description.tar.gz after 3 attempt...cription.tar.gz', '--exclude=']' returned non-zero exit status 1.") raised in repr()] LfsPath object at 0xff1da12c0d50>
dimos/utils/data.py:353: in _ensure_downloaded
    cache = get_data(filename)
        cache      = None
        filename   = 'yam_description/i2rt/yam.urdf'
        self       = <[RuntimeError("Failed to pull LFS file .../dimos/data/.lfs/yam_description.tar.gz after 3 attempt...cription.tar.gz', '--exclude=']' returned non-zero exit status 1.") raised in repr()] LfsPath object at 0xff1da12c0d50>
dimos/utils/data.py:310: in get_data
    archive_path = _decompress_archive(_pull_lfs_archive(archive_name))
        archive_name = 'yam_description'
        data_dir   = PosixPath('.../dimos/dimos/data')
        file_path  = PosixPath('.../dimos/dimos/data/yam_description/i2rt/yam.urdf')
        name       = 'yam_description/i2rt/yam.urdf'
        nested_path = PosixPath('i2rt/yam.urdf')
        path_parts = ('yam_description', 'i2rt', 'yam.urdf')
dimos/utils/data.py:254: in _pull_lfs_archive
    _lfs_pull(file_path, repo_root)
        file_path  = PosixPath('.../dimos/data/.lfs/yam_description.tar.gz')
        filename   = 'yam_description'
        repo_root  = PosixPath('.../work/dimos/dimos')
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

file_path = PosixPath('.../dimos/data/.lfs/yam_description.tar.gz')
repo_root = PosixPath('.../work/dimos/dimos')

    def _lfs_pull(file_path: Path, repo_root: Path, *, retries: int = 2) -> None:
        relative_path = file_path.relative_to(repo_root)
    
        env = os.environ.copy()
        env["GIT_LFS_FORCE_PROGRESS"] = "1"
    
        last_err: subprocess.CalledProcessError | None = None
        for attempt in range(1, retries + 2):  # retries + 1 total attempts
            try:
                subprocess.run(
                    # --exclude= overrides lfs.fetchexclude from .lfsconfig, which
                    # otherwise silently skips data/.lfs/* even when --include matches.
                    ["git", "lfs", "pull", "--include", str(relative_path), "--exclude="],
                    cwd=repo_root,
                    check=True,
                    env=env,
                )
                return
            except subprocess.CalledProcessError as e:
                last_err = e
                if attempt <= retries:
                    time.sleep(attempt)  # 1s, 2s backoff
    
>       raise RuntimeError(
            f"Failed to pull LFS file {file_path} after {retries + 1} attempts: {last_err}"
        )
E       RuntimeError: Failed to pull LFS file .../dimos/data/.lfs/yam_description.tar.gz after 3 attempts: Command '['git', 'lfs', 'pull', '--include', 'data/.lfs/yam_description.tar.gz', '--exclude=']' returned non-zero exit status 1.

attempt    = 3
env        = {'ACCEPT_EULA': 'Y', 'ACTIONS_ID_TOKEN_REQUEST_TOKEN': 'eyJhbGciOiJSUzI1NiIsImtpZCI6IjM4ODI2YjE3LTZhMzAtNWY5Yi1iMTY5LT...-version=2.0', 'ACTIONS_ORCHESTRATION_ID': '7da26a0d-a906-4b2b-9dae-d13f44e84b20.tests.ubuntu-24_04-arm_3_14_fal', ...}
file_path  = PosixPath('.../dimos/data/.lfs/yam_description.tar.gz')
last_err   = CalledProcessError(1, ['git', 'lfs', 'pull', '--include', 'data/.lfs/yam_description.tar.gz', '--exclude='])
relative_path = PosixPath('data/.lfs/yam_description.tar.gz')
repo_root  = PosixPath('.../work/dimos/dimos')
retries    = 2

dimos/utils/data.py:224: RuntimeError

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

mintlify · 2026-08-14T20:03:58Z

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated (UTC)
dimensional	🟢 Ready	View Preview	Aug 14, 2026, 8:04 PM

💡 Tip: Enable Workflows to automatically generate PRs for you.

TomCC7 · 2026-09-01T02:58:12Z

The combined change has now been split into narrow dependent PRs:

fix(openyam): align learning stack with canonical model #3853 — OpenYAM learning baseline
refactor(imitation): isolate LeRobot policy runtime #3856 — isolated LeRobot policy runtime
feat(imitation): add native OpenYAM collection #3854 — native OpenYAM collection
feat(imitation): add controlled OpenYAM policy rollout #3855 — controlled OpenYAM policy rollout

GitHub does not allow changing the base of a PR already contained in a Stack, and #3315 has unrelated A1Z work above it. This PR remains open as the review-history and A1Z integration anchor; the new PRs carry the decision-clean diffs. Existing threads here are intentionally preserved, including threads that now reference files extracted into later layers.

greptile-apps · 2026-09-01T18:49:32Z

Greptile Summary

This change adds a host-side LeRobot policy integration backed by an isolated Python runtime, including rollout lifecycle handling and joint-command publication.

Two failures were reproduced. Standard wheel installations omit the isolated runtime manifest required to start LeRobot, so rollout startup raises FileNotFoundError. Separately, stopping a rollout can still emit one arm command and one gripper command after the stop request. These issues must be fixed before merge.

Confidence Score: 2/5

Not safe to merge because wheel-installed LeRobot rollouts cannot start and stopped rollouts can still publish robot commands.

The reproduced packaging failure blocks the feature in standard deployments, while the reproduced stop race can issue commands after a stop request.

Files Needing Attention: MANIFEST.in and dimos/imitation/policy/lerobot/python/dimos_lerobot/runtime.py

T-Rex Logs

What T-Rex did

Executed the installed-wheel runtime-project execution script and captured its output to document the runtime path for the P1 finding.
Executed the deterministic stop-to-publication window test and captured the output showing commands after the stop.
Produced and documented a third finding-proof for a posted P1 finding.
Validated the general-contract scenario by running a controlled sequence where stop_requested_before_arm_publish=True and recording the subsequent one arm and one gripper command after the stop.

_{Ran code and verified through T-Rex}

Comments Outside Diff (1)

General comment

Stop request can be followed by arm and gripper command publication
- Bug
  - A deterministic controlled-hook execution reached the post-predict check with stop unset, issued _stop_policy immediately afterward, and observed an arm command plus gripper command publish while the stop event was already set.
- Cause
  - _run_rollout checks _stop_event at lines 333-336 but performs publication later at lines 347-354 without a synchronization mechanism or a second stop check that closes the intervening window.
- Fix
  - Serialize stop and publication with a shared lock/protocol, or perform a final stop-state check immediately before each publication while ensuring a concurrent stop cannot occur between that check and publish.
_{Ran code and verified through T-Rex}

_{Reviews (1): Last reviewed commit: "fix(imitation): tighten LeRobot runtime ..." | Re-trigger Greptile}

greptile-apps · 2026-09-01T18:49:35Z

+include dimos/imitation/policy/lerobot/python/pyproject.toml
+include dimos/imitation/policy/lerobot/python/uv.lock


Wheel omits isolated runtime

The wheel does not contain the isolated LeRobot runtime manifest. These MANIFEST.in entries preserve the files for source distributions, but the installed wheel omits dimos/imitation/policy/lerobot/python/pyproject.toml. The host module resolves this manifest from its installed package directory before starting the isolated runtime, so wheel installations fail with FileNotFoundError before a rollout can start. Package the isolated runtime project files in the wheel, not only the source distribution.

Artifacts

Installed-wheel runtime-project execution script

This script builds and installs the wheel outside the checkout, then executes the installed runtime-project lookup; takeaway: it directly reproduces the missing-manifest failure.

Installed-wheel runtime-project execution output

This captured command output shows the wheel and installed site-packages omit the required manifest and the actual installed lookup raises FileNotFoundError; takeaway: the claimed defect is verified.

_{Ran code and verified through T-Rex}

greptile-apps · 2026-09-01T18:49:35Z

+                if self._stop_event.is_set() or (
+                    deadline is not None and time.monotonic() >= deadline
+                ):
+                    break
+
+                gripper_name = self.config.gripper_joint_name
+                gripper_index = (
+                    self.config.joint_names.index(gripper_name)
+                    if gripper_name is not None
+                    else None
+                )
+                arm_indices = [
+                    index for index in range(len(self.config.joint_names)) if index != gripper_index
+                ]
+                self.joint_command.publish(
+                    JointState(
+                        name=[self.config.joint_names[index] for index in arm_indices],
+                        position=[float(action[index]) for index in arm_indices],
+                    )
+                )
+                if gripper_index is not None:
+                    self.gripper_command.publish(Float32(data=float(action[gripper_index])))


Stop races with command publication

A stop request can arrive immediately after the stop-state check and before either publisher is called. In that interval, the rollout emits the already-computed arm command and, when configured, the gripper command despite the stop request having completed. Serialize stopping with publication, or otherwise close this window so no command can be published after stop_rollout() returns.

Artifacts

Deterministic stop-to-publication window test source

Authored controlled-hook test imports and executes the runtime implementation, holds at the line-333 check, requests stop, and asserts that subsequent publications occur; the takeaway is that the exact race is reproducible.

Controlled stop-window test output showing commands after stop

Captured output from `python trex-artifacts/stop-publication-window-test.py` in `/home/user/repo` exits successfully and records stop before one arm and one gripper publication; the takeaway is that the claimed post-stop publication path occurs.

_{Ran code and verified through T-Rex}

TomCC7 commented Jul 31, 2026

View reviewed changes

Comment thread dimos/experimental/robot_policy/lerobot.py Outdated

Comment thread dimos/experimental/robot_policy/lerobot.py Outdated

Comment thread dimos/experimental/robot_policy/lerobot.py Outdated

TomCC7 force-pushed the feat/experimental-lerobot-policy branch from a8ab2d8 to 486d959 Compare July 31, 2026 23:35

TomCC7 changed the base branch from main to feat/galaxea-a1z-hardware July 31, 2026 23:35

TomCC7 marked this pull request as ready for review July 31, 2026 23:58

TomCC7 requested review from Dreamsorcerer, leshy, mustafab0, paul-nechifor and spomichter as code owners July 31, 2026 23:58

github-actions Bot added the ready-to-merge Required CI checks have passed on this PR label Aug 1, 2026

TomCC7 mentioned this pull request Aug 1, 2026

feat(a1z): a1z learning workflow #3318

Draft

3 tasks

Base automatically changed from feat/galaxea-a1z-hardware to main August 4, 2026 08:05

TomCC7 force-pushed the feat/experimental-lerobot-policy branch from 01aa5d0 to 18ffcc7 Compare August 14, 2026 20:03

github-actions Bot removed the ready-to-merge Required CI checks have passed on this PR label Aug 14, 2026

mintlify Bot deployed to staging - docs August 14, 2026 20:04 View deployment

TomCC7 changed the base branch from main to cc/feat/openyam-macos-support August 14, 2026 20:05

TomCC7 force-pushed the feat/experimental-lerobot-policy branch from 18ffcc7 to cda4f1f Compare August 14, 2026 21:37

mintlify Bot deployed to staging - docs August 14, 2026 21:37 View deployment

TomCC7 force-pushed the feat/experimental-lerobot-policy branch from f6a2f76 to 2a46dfb Compare August 14, 2026 21:40

mintlify Bot deployed to staging - docs August 14, 2026 21:41 View deployment

TomCC7 force-pushed the feat/experimental-lerobot-policy branch from 2a46dfb to 884a430 Compare August 14, 2026 21:56

TomCC7 marked this pull request as draft August 14, 2026 21:57

mintlify Bot deployed to staging - docs August 14, 2026 21:58 View deployment

TomCC7 mentioned this pull request Aug 15, 2026

feat: add isolated LeRobot policy module #3479

Closed

1 task

TomCC7 force-pushed the feat/experimental-lerobot-policy branch from 884a430 to 2ffca94 Compare August 18, 2026 00:07

TomCC7 force-pushed the feat/experimental-lerobot-policy branch from 1f81e28 to a341360 Compare September 1, 2026 02:55

This was referenced Sep 1, 2026

feat(imitation): add native OpenYAM collection #3854

Open

refactor(imitation): isolate LeRobot policy runtime #3856

Closed

feat(imitation): add controlled OpenYAM policy rollout #3855

Draft

fix(openyam): align learning stack with canonical model #3853

Open

TomCC7 changed the base branch from cc/feat/external-python-native-module to fix/openyam-learning-baseline September 1, 2026 03:08

TomCC7 changed the title ~~feat: LeRobot rollout on openyam~~ refactor(imitation): isolate LeRobot policy runtime Sep 1, 2026

TomCC7 commented Sep 1, 2026

View reviewed changes

Comment thread dimos/codebase_checks/test_no_init_files.py Outdated

TomCC7 commented Sep 1, 2026

View reviewed changes

Comment thread dimos/core/python_native_environment.py Outdated

TomCC7 force-pushed the feat/experimental-lerobot-policy branch from 34ff638 to 89cc9f6 Compare September 1, 2026 05:15

TomCC7 force-pushed the feat/experimental-lerobot-policy branch from 89cc9f6 to b8a7241 Compare September 1, 2026 17:00

TomCC7 force-pushed the feat/experimental-lerobot-policy branch from b8a7241 to 0c8996a Compare September 1, 2026 17:16

TomCC7 force-pushed the feat/experimental-lerobot-policy branch from 0c8996a to 564d788 Compare September 1, 2026 17:17

TomCC7 commented Sep 1, 2026

View reviewed changes

Comment thread dimos/imitation/policy/lerobot/python/dimos_lerobot/runtime.py Outdated

TomCC7 commented Sep 1, 2026

View reviewed changes

Comment thread dimos/imitation/policy/lerobot/python/dimos_lerobot/runtime.py Outdated

TomCC7 and others added 8 commits September 1, 2026 11:44

feat: add isolated LeRobot policy module

94c178f

refactor: promote LeRobot policy module

34f64b2

Delete examples/native-modules/python_lerobot.py

e0f990f

refactor: isolate LeRobot policy runtime

5746ae3

refactor: keep LeRobot policy layer self-contained

76beb42

fix(imitation): update isolated runtime imports

606179e

fix(imitation): tighten LeRobot runtime typing

ca4ac7d

test(imitation): validate typed policy config mappings

80b882f

greptile-apps Bot reviewed Sep 1, 2026

View reviewed changes

Navigation Menu

Uh oh!

refactor(imitation): isolate LeRobot policy runtime - #3315

refactor(imitation): isolate LeRobot policy runtime#3315
TomCC7 wants to merge 8 commits into
fix/openyam-learning-baselinefrom
feat/experimental-lerobot-policy

TomCC7 commented Jul 31, 2026 •

edited

Loading

Uh oh!

codecov Bot commented Jul 31, 2026 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

Uh oh!

mintlify Bot commented Aug 14, 2026 •

edited

Loading

Uh oh!

TomCC7 commented Sep 1, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

greptile-apps Bot commented Sep 1, 2026 •

edited

Loading

T-Rex Logs

Comments Outside Diff (1)

Uh oh!

greptile-apps Bot Sep 1, 2026

Uh oh!

greptile-apps Bot Sep 1, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

		include dimos/imitation/policy/lerobot/python/pyproject.toml
		include dimos/imitation/policy/lerobot/python/uv.lock

Uh oh!

Conversation

TomCC7 commented Jul 31, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Problem

Solution

Learning slice

Checklist

Uh oh!

codecov Bot commented Jul 31, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

❌ 1 Tests Failed:

Uh oh!

Uh oh!

Uh oh!

Uh oh!

mintlify Bot commented Aug 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

TomCC7 commented Sep 1, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

greptile-apps Bot commented Sep 1, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Greptile Summary

Confidence Score: 2/5

T-Rex Logs

Comments Outside Diff (1)

Uh oh!

greptile-apps Bot Sep 1, 2026

Choose a reason for hiding this comment

Uh oh!

greptile-apps Bot Sep 1, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

TomCC7 commented Jul 31, 2026 •

edited

Loading

codecov Bot commented Jul 31, 2026 •

edited

Loading

mintlify Bot commented Aug 14, 2026 •

edited

Loading

greptile-apps Bot commented Sep 1, 2026 •

edited

Loading