diff --git a/backend/operator-runner/main.py b/backend/operator-runner/main.py
index e7828701..f5c3a663 100644
--- a/backend/operator-runner/main.py
+++ b/backend/operator-runner/main.py
@@ -57,6 +57,7 @@ from backend.operators.builtin.lerobot_metadata_consistency_validate import oper
from backend.operators.builtin.language_instruction_validate import operator as language_instruction_validate_operator
from backend.operators.builtin.mcap_topic_summary import operator as mcap_topic_summary_operator
from backend.operators.builtin.merge import operator as merge_operator
+from backend.operators.builtin.mjcf_actuator_validate import operator as mjcf_actuator_validate_operator
from backend.operators.builtin.model_inference import operator as model_inference_operator
from backend.operators.builtin.mongodb_write import operator as mongodb_write_operator
from backend.operators.builtin.package import operator as package_operator
@@ -64,6 +65,7 @@ from backend.operators.builtin.parquet_table_summary import operator as parquet_
from backend.operators.builtin.pointcloud_summary import operator as pointcloud_summary_operator
from backend.operators.builtin.real_source import operator as real_source_operator
from backend.operators.builtin.rlds_episode_summary import operator as rlds_episode_summary_operator
+from backend.operators.builtin.robomimic_hdf5_validate import operator as robomimic_hdf5_validate_operator
from backend.operators.builtin.robot_model_asset_validate import operator as robot_model_asset_validate_operator
from backend.operators.builtin.rosbag_summary import operator as rosbag_summary_operator
from backend.operators.builtin.rgbd_frame_pair_validate import operator as rgbd_frame_pair_validate_operator
@@ -72,6 +74,7 @@ from backend.operators.builtin.sensor_timestamp_sync_validate import operator as
from backend.operators.builtin.shell_command import operator as shell_command_operator
from backend.operators.builtin.template_render import operator as template_render_operator
from backend.operators.builtin.tfrecord_structural_validate import operator as tfrecord_structural_validate_operator
+from backend.operators.builtin.tf_transform_tree_validate import operator as tf_transform_tree_validate_operator
from backend.operators.builtin.trajectory_integrity_validate import operator as trajectory_integrity_validate_operator
from backend.operators.builtin.upload import operator as upload_operator
from backend.operators.builtin.video_annotate import operator as video_annotate_operator
@@ -4944,6 +4947,7 @@ BUILTIN_HANDLER_RULES: tuple[tuple[str, tuple[str, ...]], ...] = (
("hf_dataset_probe", ("hf 数据集探测", "huggingface dataset", "hf dataset", "dataset viewer")),
("rlds_episode_summary", ("rlds", "tfds", "open x-embodiment", "轨迹概览")),
("aloha_hdf5_validate", ("aloha hdf5", "aloha 示教", "act hdf5", "aloha 数据校验")),
+ ("robomimic_hdf5_validate", ("robomimic hdf5", "robomimic 示教", "robomimic 数据校验", "robomimic dataset")),
("hdf5_demonstration_summary", ("hdf5 示教", "hdf5 demo", "robomimic", "droid hdf5")),
("vla_sample_manifest_validate", ("vla 样本", "vla manifest", "openvla", "视觉语言动作", "droid manifest")),
("lerobot_dataset_validate", ("lerobot 数据集校验", "lerobot validate", "lerobot validation")),
@@ -4959,6 +4963,8 @@ BUILTIN_HANDLER_RULES: tuple[tuple[str, tuple[str, ...]], ...] = (
("webdataset_shard_summary", ("webdataset 分片", "webdataset shard", "tar shard", "分片概览")),
("zarr_trajectory_summary", ("zarr 轨迹", "zarr trajectory", "zarr array", "数组轨迹")),
("camera_calibration_validate", ("相机标定校验", "camera calibration", "intrinsics", "extrinsics")),
+ ("tf_transform_tree_validate", ("tf 坐标树", "transform tree", "tf tree", "坐标树校验", "frametree")),
+ ("mjcf_actuator_validate", ("mjcf 执行器", "mjcf actuator", "mujoco actuator", "执行器校验", "ctrlrange")),
("robot_model_asset_validate", ("机器人模型资产校验", "robot model asset", "urdf", "mjcf")),
("pointcloud_summary", ("点云数据概览", "point cloud", "pointcloud", "pcd", "ply")),
("sensor_timestamp_sync_validate", ("多传感器时间同步", "timestamp sync", "sensor sync", "时间同步")),
@@ -5016,6 +5022,7 @@ BUILTIN_HANDLERS: dict[str, BuiltinHandler] = {
"rlds_episode_summary": operator_handler(rlds_episode_summary_operator),
"hdf5_demonstration_summary": operator_handler(hdf5_demonstration_summary_operator),
"aloha_hdf5_validate": operator_handler(aloha_hdf5_validate_operator),
+ "robomimic_hdf5_validate": operator_handler(robomimic_hdf5_validate_operator),
"vla_sample_manifest_validate": operator_handler(vla_sample_manifest_validate_operator),
"lerobot_dataset_validate": operator_handler(lerobot_dataset_validate_operator),
"lerobot_metadata_consistency_validate": operator_handler(lerobot_metadata_consistency_validate_operator),
@@ -5030,6 +5037,8 @@ BUILTIN_HANDLERS: dict[str, BuiltinHandler] = {
"webdataset_shard_summary": operator_handler(webdataset_shard_summary_operator),
"zarr_trajectory_summary": operator_handler(zarr_trajectory_summary_operator),
"camera_calibration_validate": operator_handler(camera_calibration_validate_operator),
+ "tf_transform_tree_validate": operator_handler(tf_transform_tree_validate_operator),
+ "mjcf_actuator_validate": operator_handler(mjcf_actuator_validate_operator),
"robot_model_asset_validate": operator_handler(robot_model_asset_validate_operator),
"pointcloud_summary": operator_handler(pointcloud_summary_operator),
"sensor_timestamp_sync_validate": operator_handler(sensor_timestamp_sync_validate_operator),
diff --git a/backend/operator-runner/test_main.py b/backend/operator-runner/test_main.py
index aefbbbb8..6a76821c 100644
--- a/backend/operator-runner/test_main.py
+++ b/backend/operator-runner/test_main.py
@@ -259,8 +259,11 @@ class RuntimeFileContractTests(unittest.TestCase):
self.assertEqual(MODULE.select_handler("BOS 增量扫描"), "bos_incremental_scan")
self.assertEqual(MODULE.select_handler("人工审核"), "human_review")
self.assertEqual(MODULE.select_handler("ALOHA HDF5 示教校验"), "aloha_hdf5_validate")
+ self.assertEqual(MODULE.select_handler("robomimic HDF5 示教校验"), "robomimic_hdf5_validate")
self.assertEqual(MODULE.select_handler("VLA 样本清单校验"), "vla_sample_manifest_validate")
self.assertEqual(MODULE.select_handler("URDF 关节限位校验"), "joint_limit_validate")
+ self.assertEqual(MODULE.select_handler("MJCF 执行器校验"), "mjcf_actuator_validate")
+ self.assertEqual(MODULE.select_handler("TF 坐标树校验"), "tf_transform_tree_validate")
def test_operator_helpers_do_not_expose_legacy_builtin_handlers(self):
legacy_keys = sorted(key for key in MODULE.OPERATOR_HELPERS if key.startswith("builtin_"))
@@ -2905,6 +2908,55 @@ class EmbodiedDatasetOperatorTests(unittest.TestCase):
self.assertEqual(payload["qposDimensions"], [14])
self.assertEqual(payload["files"][0]["cameraNames"], ["cam_high", "cam_wrist"])
+ def test_robomimic_hdf5_validate_reads_real_demo_groups(self):
+ try:
+ import h5py # type: ignore
+ import numpy as np # type: ignore
+ except Exception as exc:
+ self.skipTest(f"h5py/numpy not available: {exc}")
+ with tempfile.TemporaryDirectory() as temp_dir:
+ path = pathlib.Path(temp_dir) / "robomimic.hdf5"
+ with h5py.File(path, "w") as handle:
+ data = handle.create_group("data")
+ data.attrs["total"] = 3
+ data.attrs["env_args"] = json.dumps({"env_name": "Lift"})
+ demo_1 = data.create_group("demo_1")
+ demo_1.attrs["num_samples"] = 2
+ demo_1.create_dataset("actions", data=np.array([[0.1, 0.2], [0.3, 0.4]], dtype="float32"))
+ demo_1.create_dataset("rewards", data=np.array([0.0, 1.0], dtype="float32"))
+ demo_1.create_dataset("dones", data=np.array([False, True], dtype="bool"))
+ demo_1.create_dataset("states", data=np.array([[1.0], [2.0]], dtype="float32"))
+ obs = demo_1.create_group("obs")
+ obs.create_dataset("robot0_eef_pos", data=np.array([[0.0, 0.1, 0.2], [0.2, 0.3, 0.4]], dtype="float32"))
+ next_obs = demo_1.create_group("next_obs")
+ next_obs.create_dataset("robot0_eef_pos", data=np.array([[0.1, 0.2, 0.3], [0.3, 0.4, 0.5]], dtype="float32"))
+ demo_2 = data.create_group("demo_2")
+ demo_2.attrs["num_samples"] = 1
+ demo_2.create_dataset("actions", data=np.array([[0.5, 0.6]], dtype="float32"))
+ demo_2.create_dataset("rewards", data=np.array([0.5], dtype="float32"))
+ demo_2.create_dataset("dones", data=np.array([True], dtype="bool"))
+ obs = demo_2.create_group("obs")
+ obs.create_dataset("robot0_eef_pos", data=np.array([[0.4, 0.5, 0.6]], dtype="float32"))
+ next_obs = demo_2.create_group("next_obs")
+ next_obs.create_dataset("robot0_eef_pos", data=np.array([[0.5, 0.6, 0.7]], dtype="float32"))
+ mask = handle.create_group("mask")
+ mask.create_dataset("train", data=np.array([b"demo_1"]))
+ mask.create_dataset("valid", data=np.array([b"demo_2"]))
+ node_data = {"label": "robomimic HDF5 示教校验", "hyperparams": [{"key": "SOURCE_PATH", "value": str(path)}]}
+
+ artifacts = MODULE.BUILTIN_HANDLERS["robomimic_hdf5_validate"](
+ "run-1", "robomimic", pathlib.Path(temp_dir) / "out", node_data, []
+ )
+
+ payload = MODULE.load_artifact_payload(artifacts[0])
+ self.assertEqual(payload["datasetType"], "robomimic-hdf5-validation")
+ self.assertTrue(payload["valid"])
+ self.assertEqual(payload["demoCount"], 2)
+ self.assertEqual(payload["totalSamples"], 3)
+ self.assertEqual(payload["maskKeys"], ["train", "valid"])
+ self.assertEqual(payload["actionDimensions"], [2])
+ self.assertEqual(payload["files"][0]["observationKeyCounts"], {"robot0_eef_pos": 2})
+
def test_lerobot_dataset_validate_reads_real_parquet_and_metadata(self):
try:
import pyarrow as pa # type: ignore
@@ -3484,6 +3536,79 @@ class EmbodiedDatasetOperatorTests(unittest.TestCase):
self.assertEqual(payload["meshReferenceCount"], 1)
self.assertEqual(payload["missingMeshCount"], 0)
+ def test_mjcf_actuator_validate_reads_real_ctrlrange(self):
+ with tempfile.TemporaryDirectory() as temp_dir:
+ root = pathlib.Path(temp_dir) / "mjcf"
+ root.mkdir()
+ (root / "scene.xml").write_text(
+ ""
+ " "
+ " "
+ " "
+ " "
+ " "
+ " "
+ " "
+ " "
+ " "
+ " "
+ "",
+ encoding="utf-8",
+ )
+ node_data = {"label": "MJCF 执行器校验", "hyperparams": [{"key": "SOURCE_PATH", "value": str(root)}]}
+
+ artifacts = MODULE.BUILTIN_HANDLERS["mjcf_actuator_validate"](
+ "run-1", "mjcf-actuator", root / "out", node_data, []
+ )
+
+ payload = MODULE.load_artifact_payload(artifacts[0])
+ self.assertEqual(payload["datasetType"], "mjcf-actuator-validation")
+ self.assertTrue(payload["valid"])
+ self.assertEqual(payload["modelCount"], 1)
+ self.assertEqual(payload["actuatorCount"], 2)
+ self.assertEqual(payload["invalidActuatorCount"], 0)
+ self.assertEqual(payload["issueCount"], 0)
+ self.assertEqual(payload["models"][0]["targetCounts"]["joints"], 1)
+ self.assertEqual(payload["models"][0]["targetCounts"]["sites"], 1)
+
+ def test_tf_transform_tree_validate_reads_real_transform_jsonl(self):
+ with tempfile.TemporaryDirectory() as temp_dir:
+ root = pathlib.Path(temp_dir) / "tf"
+ root.mkdir()
+ rows = [
+ {
+ "header": {"frame_id": "world"},
+ "child_frame_id": "base_link",
+ "transform": {
+ "translation": {"x": 0.0, "y": 0.0, "z": 0.0},
+ "rotation": {"x": 0.0, "y": 0.0, "z": 0.0, "w": 1.0},
+ },
+ },
+ {
+ "header": {"frame_id": "base_link"},
+ "child_frame_id": "camera",
+ "transform": {
+ "translation": {"x": 0.1, "y": 0.2, "z": 0.3},
+ "rotation": {"x": 0.0, "y": 0.0, "z": 0.70710678, "w": 0.70710678},
+ },
+ },
+ ]
+ (root / "transforms.jsonl").write_text("\n".join(json.dumps(row) for row in rows), encoding="utf-8")
+ node_data = {"label": "TF 坐标树校验", "hyperparams": [{"key": "SOURCE_PATH", "value": str(root)}]}
+
+ artifacts = MODULE.BUILTIN_HANDLERS["tf_transform_tree_validate"](
+ "run-1", "tf-tree", root / "out", node_data, []
+ )
+
+ payload = MODULE.load_artifact_payload(artifacts[0])
+ self.assertEqual(payload["datasetType"], "tf-transform-tree-validation")
+ self.assertTrue(payload["valid"])
+ self.assertEqual(payload["transformCount"], 2)
+ self.assertEqual(payload["frameCount"], 3)
+ self.assertEqual(payload["rootFrames"], ["world"])
+ self.assertEqual(payload["cycleCount"], 0)
+ self.assertEqual(payload["issueCount"], 0)
+
def test_joint_limit_validate_reads_real_urdf_and_parquet_positions(self):
try:
import pyarrow as pa # type: ignore
diff --git a/backend/operators/builtin/mjcf_actuator_validate/__init__.py b/backend/operators/builtin/mjcf_actuator_validate/__init__.py
new file mode 100644
index 00000000..8b137891
--- /dev/null
+++ b/backend/operators/builtin/mjcf_actuator_validate/__init__.py
@@ -0,0 +1 @@
+
diff --git a/backend/operators/builtin/mjcf_actuator_validate/operator.json b/backend/operators/builtin/mjcf_actuator_validate/operator.json
new file mode 100644
index 00000000..2b5f7704
--- /dev/null
+++ b/backend/operators/builtin/mjcf_actuator_validate/operator.json
@@ -0,0 +1,16 @@
+{
+ "id": "operator-builtin-mjcf-actuator-validate",
+ "name": "MJCF 执行器校验",
+ "category": "ROBOTICS",
+ "type": "builtin",
+ "runtimeHandler": "builtin_mjcf_actuator_validate",
+ "runtimeSource": "backend/operators/builtin/mjcf_actuator_validate/operator.py",
+ "migrationStatus": "migrated",
+ "acceptance": {
+ "requiresRealInputs": true,
+ "mockOnlyAccepted": false,
+ "simulationOnlyAccepted": false,
+ "hardcodedOutputAccepted": false,
+ "shallowOnlyAccepted": false
+ }
+}
diff --git a/backend/operators/builtin/mjcf_actuator_validate/operator.py b/backend/operators/builtin/mjcf_actuator_validate/operator.py
new file mode 100644
index 00000000..1a5e2591
--- /dev/null
+++ b/backend/operators/builtin/mjcf_actuator_validate/operator.py
@@ -0,0 +1,226 @@
+from __future__ import annotations
+
+import math
+import pathlib
+import xml.etree.ElementTree as ET
+from typing import Any, Callable
+
+
+HelperMap = dict[str, Callable[..., Any]]
+MJCF_SUFFIXES = {".xml", ".mjcf"}
+ACTUATOR_TAGS = {"general", "motor", "position", "velocity", "intvelocity", "damper", "cylinder", "muscle", "adhesion"}
+TRUTHY = {"true", "1", "yes", "auto"}
+
+
+def _int_param(params: dict[str, str], key: str, default: int) -> int:
+ raw = str(params.get(key, str(default)) or str(default)).strip()
+ try:
+ value = int(raw)
+ except ValueError as exc:
+ raise RuntimeError(f"{key} must be an integer") from exc
+ if value <= 0:
+ raise RuntimeError(f"{key} must be > 0")
+ return value
+
+
+def _candidate_model_files(source_path: pathlib.Path, max_models: int) -> list[pathlib.Path]:
+ if source_path.is_file():
+ return [source_path] if source_path.suffix.lower() in MJCF_SUFFIXES else []
+ preferred = [
+ source_path / "scene.xml",
+ source_path / "model.xml",
+ source_path / "robot.xml",
+ source_path / "model.mjcf",
+ ]
+ files = [item for item in preferred if item.is_file()]
+ files.extend(
+ item
+ for item in sorted(source_path.rglob("*"))
+ if item.is_file() and item.suffix.lower() in MJCF_SUFFIXES and item not in files
+ )
+ return files[:max_models]
+
+
+def _strip_tag(tag: str) -> str:
+ return tag.rsplit("}", 1)[-1] if "}" in tag else tag
+
+
+def _number_list(value: str) -> list[float]:
+ numbers: list[float] = []
+ for item in value.replace(",", " ").split():
+ try:
+ number = float(item)
+ except ValueError:
+ return []
+ if not math.isfinite(number):
+ return []
+ numbers.append(number)
+ return numbers
+
+
+def _validate_range(raw: str, key: str, issues: list[dict[str, Any]], actuator_name: str) -> list[float]:
+ numbers = _number_list(raw)
+ if len(numbers) != 2:
+ issues.append({"code": "invalid_range", "actuator": actuator_name, "attribute": key, "value": raw})
+ return []
+ if numbers[0] >= numbers[1]:
+ issues.append({"code": "non_increasing_range", "actuator": actuator_name, "attribute": key, "value": numbers})
+ return numbers
+
+
+def _named_elements(root: ET.Element, tags: set[str]) -> set[str]:
+ names: set[str] = set()
+ for element in root.iter():
+ if _strip_tag(element.tag) in tags:
+ name = str(element.attrib.get("name", "")).strip()
+ if name:
+ names.add(name)
+ return names
+
+
+def _actuator_elements(root: ET.Element) -> list[ET.Element]:
+ parent_map = {child: parent for parent in root.iter() for child in parent}
+ actuators: list[ET.Element] = []
+ for element in root.iter():
+ if _strip_tag(element.tag) not in ACTUATOR_TAGS:
+ continue
+ parent = parent_map.get(element)
+ if parent is not None and _strip_tag(parent.tag) == "actuator":
+ actuators.append(element)
+ return actuators
+
+
+def _validate_actuator(index: int, element: ET.Element, references: dict[str, set[str]], seen_names: set[str]) -> dict[str, Any]:
+ tag = _strip_tag(element.tag)
+ name = str(element.attrib.get("name", "") or f"{tag}-{index}").strip()
+ issues: list[dict[str, Any]] = []
+ if element.attrib.get("name"):
+ if name in seen_names:
+ issues.append({"code": "duplicate_actuator_name", "actuator": name})
+ seen_names.add(name)
+ target_refs: dict[str, str] = {}
+ for attr, ref_set in (
+ ("joint", references["joints"]),
+ ("tendon", references["tendons"]),
+ ("site", references["sites"]),
+ ("body", references["bodies"]),
+ ("cranksite", references["sites"]),
+ ("slidersite", references["sites"]),
+ ("refsite", references["sites"]),
+ ):
+ raw = str(element.attrib.get(attr, "")).strip()
+ if not raw:
+ continue
+ target_refs[attr] = raw
+ if raw not in ref_set:
+ issues.append({"code": "missing_actuator_target", "actuator": name, "attribute": attr, "target": raw})
+ if not target_refs and tag != "adhesion":
+ issues.append({"code": "missing_actuator_target", "actuator": name})
+
+ ranges: dict[str, list[float]] = {}
+ for attr in ("ctrlrange", "forcerange", "actrange"):
+ if attr in element.attrib:
+ ranges[attr] = _validate_range(str(element.attrib[attr]), attr, issues, name)
+ for limited_attr, range_attr in (("ctrllimited", "ctrlrange"), ("forcelimited", "forcerange"), ("actlimited", "actrange")):
+ raw = str(element.attrib.get(limited_attr, "")).strip().lower()
+ if raw in TRUTHY and range_attr not in ranges:
+ issues.append({"code": "limited_without_range", "actuator": name, "limited": limited_attr, "range": range_attr})
+ if "gear" in element.attrib and not _number_list(str(element.attrib["gear"])):
+ issues.append({"code": "invalid_gear_vector", "actuator": name, "gear": str(element.attrib["gear"])})
+
+ return {
+ "name": name,
+ "type": tag,
+ "valid": not issues,
+ "issueCount": len(issues),
+ "issues": issues,
+ "targets": target_refs,
+ "ranges": ranges,
+ "gear": _number_list(str(element.attrib.get("gear", ""))) if element.attrib.get("gear") else [],
+ }
+
+
+def _summarize_model(path: pathlib.Path) -> dict[str, Any]:
+ try:
+ tree = ET.parse(path)
+ except ET.ParseError as exc:
+ raise RuntimeError(f"{path} is not valid MJCF XML: {exc}") from exc
+ root = tree.getroot()
+ root_tag = _strip_tag(root.tag)
+ model_issues: list[dict[str, Any]] = []
+ if root_tag != "mujoco":
+ model_issues.append({"code": "not_mjcf_root", "rootTag": root_tag})
+ references = {
+ "joints": _named_elements(root, {"joint"}),
+ "tendons": _named_elements(root, {"fixed", "spatial"}),
+ "sites": _named_elements(root, {"site"}),
+ "bodies": _named_elements(root, {"body"}),
+ }
+ seen_names: set[str] = set()
+ actuators = [_validate_actuator(index, element, references, seen_names) for index, element in enumerate(_actuator_elements(root))]
+ issues = list(model_issues)
+ issues.extend({"actuator": actuator["name"], **issue} for actuator in actuators for issue in actuator["issues"])
+ if root_tag == "mujoco" and not actuators:
+ issues.append({"code": "missing_actuator_elements"})
+ return {
+ "path": str(path),
+ "valid": not issues,
+ "issueCount": len(issues),
+ "issues": issues,
+ "rootTag": root_tag,
+ "jointCount": len(references["joints"]),
+ "tendonCount": len(references["tendons"]),
+ "siteCount": len(references["sites"]),
+ "bodyCount": len(references["bodies"]),
+ "actuatorCount": len(actuators),
+ "invalidActuatorCount": sum(1 for item in actuators if not item["valid"]),
+ "targetCounts": {key: len(value) for key, value in references.items()},
+ "actuators": actuators,
+ }
+
+
+def run(
+ run_id: str,
+ node_id: str,
+ output_dir: Any,
+ node_data: dict[str, Any],
+ inputs: list[dict[str, Any]],
+ helpers: HelperMap,
+) -> list[dict[str, Any]]:
+ params = helpers["hyperparams"](node_data)
+ source_path = helpers["resolve_source_path"](params, inputs)
+ if source_path is None:
+ raise RuntimeError("MJCF 执行器校验需要 SOURCE_PATH 指向真实 MJCF XML 文件或目录")
+ max_models = _int_param(params, "MAX_MODELS", 20)
+ models = _candidate_model_files(source_path, max_models)
+ if not models:
+ raise RuntimeError(f"{source_path} does not contain MJCF XML files")
+ summaries = [_summarize_model(path) for path in models]
+ issues = [
+ {"model": item["path"], **issue}
+ for item in summaries
+ for issue in item.get("issues", [])
+ ]
+ payload = {
+ "datasetType": "mjcf-actuator-validation",
+ "sourcePath": str(source_path),
+ "valid": not issues,
+ "issueCount": len(issues),
+ "issues": issues[:300],
+ "modelCount": len(summaries),
+ "actuatorCount": sum(int(item.get("actuatorCount", 0)) for item in summaries),
+ "invalidActuatorCount": sum(int(item.get("invalidActuatorCount", 0)) for item in summaries),
+ "jointCount": sum(int(item.get("jointCount", 0)) for item in summaries),
+ "siteCount": sum(int(item.get("siteCount", 0)) for item in summaries),
+ "models": summaries,
+ }
+ return [
+ helpers["write_json_artifact"](
+ run_id,
+ node_id,
+ output_dir,
+ "mjcf-actuator-validation.json",
+ payload,
+ output=helpers["output_binding"](node_data),
+ )
+ ]
diff --git a/backend/operators/builtin/mjcf_actuator_validate/repository.json b/backend/operators/builtin/mjcf_actuator_validate/repository.json
new file mode 100644
index 00000000..40a4645a
--- /dev/null
+++ b/backend/operators/builtin/mjcf_actuator_validate/repository.json
@@ -0,0 +1,31 @@
+{
+ "acceptance": {
+ "hardcodedOutputAccepted": false,
+ "mockOnlyAccepted": false,
+ "pocFullAcceptanceScript": "scripts/poc-operator-full-acceptance.sh",
+ "requiresPersistentWorkspaceEvidence": true,
+ "requiresRealInputs": true,
+ "shallowOnlyAccepted": false,
+ "simulationOnlyAccepted": false
+ },
+ "category": "ROBOTICS",
+ "containerEntrypoint": "backend/operator-runner/main.py",
+ "entrypoint": "operator.py",
+ "metadata": "operator.json",
+ "name": "MJCF 执行器校验",
+ "operatorId": "operator-builtin-mjcf-actuator-validate",
+ "operatorSlug": "mjcf_actuator_validate",
+ "preferredSlug": "mjcf_actuator_validate",
+ "pythonImport": "backend.operators.builtin.mjcf_actuator_validate.operator:run",
+ "repositoryId": "builtin/mjcf_actuator_validate",
+ "runtimeDependencies": [
+ "mjcf-files",
+ "xml-parser"
+ ],
+ "runtimeHandler": "builtin_mjcf_actuator_validate",
+ "runtimeSource": "backend/operators/builtin/mjcf_actuator_validate/operator.py",
+ "schemaVersion": "droboticflow.operator.repository/v1",
+ "sourceRoot": "backend/operators/builtin/mjcf_actuator_validate",
+ "strictPocAcceptanceExternals": [],
+ "type": "builtin"
+}
diff --git a/backend/operators/builtin/robomimic_hdf5_validate/__init__.py b/backend/operators/builtin/robomimic_hdf5_validate/__init__.py
new file mode 100644
index 00000000..8b137891
--- /dev/null
+++ b/backend/operators/builtin/robomimic_hdf5_validate/__init__.py
@@ -0,0 +1 @@
+
diff --git a/backend/operators/builtin/robomimic_hdf5_validate/operator.json b/backend/operators/builtin/robomimic_hdf5_validate/operator.json
new file mode 100644
index 00000000..9e186c3f
--- /dev/null
+++ b/backend/operators/builtin/robomimic_hdf5_validate/operator.json
@@ -0,0 +1,16 @@
+{
+ "id": "operator-builtin-robomimic-hdf5-validate",
+ "name": "robomimic HDF5 示教校验",
+ "category": "ROBOTICS",
+ "type": "builtin",
+ "runtimeHandler": "builtin_robomimic_hdf5_validate",
+ "runtimeSource": "backend/operators/builtin/robomimic_hdf5_validate/operator.py",
+ "migrationStatus": "migrated",
+ "acceptance": {
+ "requiresRealInputs": true,
+ "mockOnlyAccepted": false,
+ "simulationOnlyAccepted": false,
+ "hardcodedOutputAccepted": false,
+ "shallowOnlyAccepted": false
+ }
+}
diff --git a/backend/operators/builtin/robomimic_hdf5_validate/operator.py b/backend/operators/builtin/robomimic_hdf5_validate/operator.py
new file mode 100644
index 00000000..8e3998f0
--- /dev/null
+++ b/backend/operators/builtin/robomimic_hdf5_validate/operator.py
@@ -0,0 +1,372 @@
+from __future__ import annotations
+
+import json
+import math
+import pathlib
+from typing import Any, Callable
+
+
+HelperMap = dict[str, Callable[..., Any]]
+
+
+def _require_h5py():
+ try:
+ import h5py # type: ignore
+ except Exception as exc: # pragma: no cover - runtime dependency check
+ raise RuntimeError("robomimic HDF5 校验需要 h5py 运行时依赖") from exc
+ return h5py
+
+
+def _int_param(params: dict[str, str], key: str, default: int) -> int:
+ raw = str(params.get(key, str(default)) or str(default)).strip()
+ try:
+ value = int(raw)
+ except ValueError as exc:
+ raise RuntimeError(f"{key} must be an integer") from exc
+ if value <= 0:
+ raise RuntimeError(f"{key} must be > 0")
+ return value
+
+
+def _float_param(params: dict[str, str], key: str, default: float) -> float:
+ raw = str(params.get(key, str(default)) or str(default)).strip()
+ try:
+ value = float(raw)
+ except ValueError as exc:
+ raise RuntimeError(f"{key} must be a number") from exc
+ if not math.isfinite(value) or value <= 0:
+ raise RuntimeError(f"{key} must be a positive finite number")
+ return value
+
+
+def _bool_param(params: dict[str, str], key: str, default: bool) -> bool:
+ raw = str(params.get(key, str(default)) or str(default)).strip().lower()
+ if raw in {"1", "true", "yes", "y", "on"}:
+ return True
+ if raw in {"0", "false", "no", "n", "off"}:
+ return False
+ return default
+
+
+def _find_hdf5_files(source_path: pathlib.Path, max_files: int) -> list[pathlib.Path]:
+ if source_path.is_file():
+ return [source_path] if source_path.suffix.lower() in {".hdf5", ".h5"} else []
+ preferred = [source_path / "low_dim.hdf5", source_path / "image.hdf5", source_path / "demo.hdf5"]
+ files = [item for item in preferred if item.is_file()]
+ files.extend(
+ item
+ for item in sorted(source_path.rglob("*"))
+ if item.is_file() and item.suffix.lower() in {".hdf5", ".h5"} and item not in files
+ )
+ return files[:max_files]
+
+
+def _jsonable(value: Any) -> Any:
+ if isinstance(value, bytes):
+ return value.decode("utf-8", errors="replace")
+ if isinstance(value, (str, int, float, bool)) or value is None:
+ return value
+ if hasattr(value, "item"):
+ try:
+ return _jsonable(value.item())
+ except Exception:
+ pass
+ if isinstance(value, dict):
+ return {str(key): _jsonable(item) for key, item in value.items()}
+ if isinstance(value, (list, tuple)):
+ return [_jsonable(item) for item in value]
+ return str(value)
+
+
+def _attrs_payload(node: Any) -> dict[str, Any]:
+ payload: dict[str, Any] = {}
+ for key, value in node.attrs.items():
+ parsed = _jsonable(value)
+ if isinstance(parsed, str) and parsed.strip().startswith(("{", "[")):
+ try:
+ parsed = json.loads(parsed)
+ except json.JSONDecodeError:
+ pass
+ payload[str(key)] = parsed
+ return payload
+
+
+def _shape(node: Any | None) -> list[int]:
+ if node is None or not hasattr(node, "shape"):
+ return []
+ return [int(item) for item in node.shape]
+
+
+def _first_dim(node: Any | None) -> int:
+ shape = _shape(node)
+ return int(shape[0]) if shape else 0
+
+
+def _demo_group(root: Any) -> Any | None:
+ if "data" in root and hasattr(root["data"], "keys"):
+ return root["data"]
+ return root if hasattr(root, "keys") else None
+
+
+def _demo_keys(data_group: Any) -> list[str]:
+ keys: list[str] = []
+ for key in sorted(data_group.keys()):
+ item = data_group[key]
+ if hasattr(item, "keys") and (key.startswith("demo") or "actions" in item):
+ keys.append(str(key))
+ return keys
+
+
+def _iter_dataset_paths(group: Any, prefix: str = "") -> list[tuple[str, Any]]:
+ datasets: list[tuple[str, Any]] = []
+ if not hasattr(group, "keys"):
+ return datasets
+ for key in sorted(group.keys()):
+ item = group[key]
+ path = f"{prefix}/{key}" if prefix else str(key)
+ if hasattr(item, "shape") and hasattr(item, "dtype"):
+ datasets.append((path, item))
+ else:
+ datasets.extend(_iter_dataset_paths(item, path))
+ return datasets
+
+
+def _dataset_values(dataset: Any) -> list[str]:
+ raw = dataset[()]
+ if hasattr(raw, "tolist"):
+ raw = raw.tolist()
+ if not isinstance(raw, list):
+ raw = [raw]
+ return [str(_jsonable(item)) for item in raw]
+
+
+def _action_range_issue(actions: Any | None, max_abs_action: float) -> dict[str, Any] | None:
+ if actions is None or not hasattr(actions, "__getitem__"):
+ return None
+ try:
+ values = actions[()]
+ if hasattr(values, "tolist"):
+ values = values.tolist()
+ except Exception:
+ return {"code": "unreadable_actions"}
+ stack = list(values) if isinstance(values, list) else [values]
+ max_value = 0.0
+ while stack:
+ item = stack.pop()
+ if isinstance(item, list):
+ stack.extend(item)
+ elif isinstance(item, (int, float)):
+ max_value = max(max_value, abs(float(item)))
+ if max_value > max_abs_action:
+ return {"code": "action_value_out_of_expected_range", "maxAbsAction": max_value, "limit": max_abs_action}
+ return None
+
+
+def _validate_frame_dataset(
+ issues: list[dict[str, Any]],
+ dataset: Any | None,
+ dataset_path: str,
+ expected_frames: int,
+ required: bool,
+) -> list[int]:
+ shape = _shape(dataset)
+ if not shape:
+ if required:
+ issues.append({"code": "missing_or_scalar_dataset", "dataset": dataset_path})
+ return []
+ if expected_frames > 0 and int(shape[0]) != expected_frames:
+ issues.append(
+ {
+ "code": "frame_count_mismatch",
+ "dataset": dataset_path,
+ "expected": expected_frames,
+ "actual": int(shape[0]),
+ }
+ )
+ return shape
+
+
+def _summarize_demo(
+ demo_id: str,
+ group: Any,
+ *,
+ require_rewards: bool,
+ require_dones: bool,
+ require_next_obs: bool,
+ check_action_range: bool,
+ max_abs_action: float,
+) -> dict[str, Any]:
+ actions = group.get("actions") if hasattr(group, "get") else None
+ rewards = group.get("rewards") if hasattr(group, "get") else None
+ dones = group.get("dones") if hasattr(group, "get") else None
+ states = group.get("states") if hasattr(group, "get") else None
+ obs = group.get("obs") if hasattr(group, "get") else None
+ next_obs = group.get("next_obs") if hasattr(group, "get") else None
+ actions_shape = _shape(actions)
+ fallback_frames = _first_dim(actions) or _first_dim(states)
+ frame_count = int(group.attrs.get("num_samples", 0) or fallback_frames)
+ issues: list[dict[str, Any]] = []
+ if frame_count <= 0:
+ issues.append({"code": "missing_num_samples_or_frame_dimension", "demo": demo_id})
+
+ _validate_frame_dataset(issues, actions, "actions", frame_count, required=True)
+ _validate_frame_dataset(issues, rewards, "rewards", frame_count, required=require_rewards)
+ _validate_frame_dataset(issues, dones, "dones", frame_count, required=require_dones)
+ _validate_frame_dataset(issues, states, "states", frame_count, required=False)
+
+ obs_datasets = _iter_dataset_paths(obs)
+ next_obs_datasets = _iter_dataset_paths(next_obs)
+ if not obs_datasets:
+ issues.append({"code": "missing_observation_group", "group": "obs"})
+ for path, dataset in obs_datasets:
+ _validate_frame_dataset(issues, dataset, f"obs/{path}", frame_count, required=True)
+ if require_next_obs and not next_obs_datasets:
+ issues.append({"code": "missing_next_observation_group", "group": "next_obs"})
+ for path, dataset in next_obs_datasets:
+ _validate_frame_dataset(issues, dataset, f"next_obs/{path}", frame_count, required=True)
+
+ if check_action_range:
+ issue = _action_range_issue(actions, max_abs_action)
+ if issue:
+ issues.append(issue)
+
+ return {
+ "id": demo_id,
+ "valid": not issues,
+ "issueCount": len(issues),
+ "issues": issues,
+ "frameCount": frame_count,
+ "attrs": _attrs_payload(group),
+ "actionsShape": actions_shape,
+ "statesShape": _shape(states),
+ "rewardsShape": _shape(rewards),
+ "donesShape": _shape(dones),
+ "actionDimension": int(actions_shape[1]) if len(actions_shape) > 1 else 0,
+ "observationKeys": [path for path, _ in obs_datasets],
+ "nextObservationKeys": [path for path, _ in next_obs_datasets],
+ }
+
+
+def _summarize_file(path: pathlib.Path, params: dict[str, str]) -> dict[str, Any]:
+ h5py = _require_h5py()
+ max_demos = _int_param(params, "MAX_DEMOS", 200)
+ require_rewards = _bool_param(params, "REQUIRE_REWARDS", True)
+ require_dones = _bool_param(params, "REQUIRE_DONES", True)
+ require_next_obs = _bool_param(params, "REQUIRE_NEXT_OBS", True)
+ check_action_range = _bool_param(params, "CHECK_ACTION_RANGE", False)
+ max_abs_action = _float_param(params, "MAX_ABS_ACTION", 1.0)
+
+ with h5py.File(path, "r") as handle:
+ data_group = _demo_group(handle)
+ if data_group is None:
+ raise RuntimeError(f"{path} does not contain an HDF5 group tree")
+ demo_keys = _demo_keys(data_group)
+ demos = [
+ _summarize_demo(
+ demo_id,
+ data_group[demo_id],
+ require_rewards=require_rewards,
+ require_dones=require_dones,
+ require_next_obs=require_next_obs,
+ check_action_range=check_action_range,
+ max_abs_action=max_abs_action,
+ )
+ for demo_id in demo_keys[:max_demos]
+ ]
+ issues: list[dict[str, Any]] = []
+ if not demo_keys:
+ issues.append({"code": "missing_demo_groups", "path": "/data/demo_*"})
+ for demo in demos:
+ issues.extend({"demo": demo["id"], **issue} for issue in demo["issues"])
+ total_samples = sum(int(item["frameCount"]) for item in demos)
+ declared_total = int(data_group.attrs.get("total", 0) or 0)
+ if declared_total and declared_total != total_samples:
+ issues.append(
+ {
+ "code": "declared_total_mismatch",
+ "declaredTotal": declared_total,
+ "actualTotal": total_samples,
+ }
+ )
+
+ mask_keys: list[str] = []
+ mask_entries: dict[str, list[str]] = {}
+ if "mask" in handle and hasattr(handle["mask"], "keys"):
+ for key in sorted(handle["mask"].keys()):
+ dataset = handle["mask"][key]
+ if hasattr(dataset, "shape"):
+ refs = _dataset_values(dataset)
+ mask_keys.append(str(key))
+ mask_entries[str(key)] = refs
+ missing = sorted(ref for ref in refs if ref not in demo_keys)
+ if missing:
+ issues.append({"code": "mask_references_missing_demos", "mask": str(key), "missing": missing})
+
+ return {
+ "path": str(path),
+ "fileSizeBytes": path.stat().st_size,
+ "rootAttrs": _attrs_payload(handle),
+ "dataAttrs": _attrs_payload(data_group),
+ "valid": not issues,
+ "issueCount": len(issues),
+ "issues": issues,
+ "demoCount": len(demo_keys),
+ "returnedDemoCount": len(demos),
+ "totalSamples": total_samples,
+ "declaredTotal": declared_total,
+ "maskKeys": mask_keys,
+ "maskEntries": mask_entries,
+ "actionDimensions": sorted({int(item["actionDimension"]) for item in demos if item.get("actionDimension")}),
+ "observationKeyCounts": {
+ key: sum(1 for demo in demos if key in demo.get("observationKeys", []))
+ for key in sorted({key for demo in demos for key in demo.get("observationKeys", [])})
+ },
+ "demos": demos,
+ }
+
+
+def run(
+ run_id: str,
+ node_id: str,
+ output_dir: Any,
+ node_data: dict[str, Any],
+ inputs: list[dict[str, Any]],
+ helpers: HelperMap,
+) -> list[dict[str, Any]]:
+ params = helpers["hyperparams"](node_data)
+ source_path = helpers["resolve_source_path"](params, inputs)
+ if source_path is None:
+ raise RuntimeError("robomimic HDF5 校验需要 SOURCE_PATH 指向真实 .hdf5/.h5 文件或目录")
+ max_files = _int_param(params, "MAX_FILES", 20)
+ files = _find_hdf5_files(source_path, max_files)
+ if not files:
+ raise RuntimeError(f"{source_path} does not contain robomimic .hdf5/.h5 files")
+ file_summaries = [_summarize_file(path, params) for path in files]
+ issues = [
+ {"file": item["path"], **issue}
+ for item in file_summaries
+ for issue in item.get("issues", [])
+ ]
+ payload = {
+ "datasetType": "robomimic-hdf5-validation",
+ "sourcePath": str(source_path),
+ "valid": not issues,
+ "issueCount": len(issues),
+ "issues": issues[:300],
+ "fileCount": len(file_summaries),
+ "demoCount": sum(int(item.get("demoCount", 0)) for item in file_summaries),
+ "totalSamples": sum(int(item.get("totalSamples", 0)) for item in file_summaries),
+ "maskKeys": sorted({key for item in file_summaries for key in item.get("maskKeys", [])}),
+ "actionDimensions": sorted({dim for item in file_summaries for dim in item.get("actionDimensions", [])}),
+ "files": file_summaries,
+ }
+ return [
+ helpers["write_json_artifact"](
+ run_id,
+ node_id,
+ output_dir,
+ "robomimic-hdf5-validation.json",
+ payload,
+ output=helpers["output_binding"](node_data),
+ )
+ ]
diff --git a/backend/operators/builtin/robomimic_hdf5_validate/repository.json b/backend/operators/builtin/robomimic_hdf5_validate/repository.json
new file mode 100644
index 00000000..5ad86dea
--- /dev/null
+++ b/backend/operators/builtin/robomimic_hdf5_validate/repository.json
@@ -0,0 +1,32 @@
+{
+ "acceptance": {
+ "hardcodedOutputAccepted": false,
+ "mockOnlyAccepted": false,
+ "pocFullAcceptanceScript": "scripts/poc-operator-full-acceptance.sh",
+ "requiresPersistentWorkspaceEvidence": true,
+ "requiresRealInputs": true,
+ "shallowOnlyAccepted": false,
+ "simulationOnlyAccepted": false
+ },
+ "category": "ROBOTICS",
+ "containerEntrypoint": "backend/operator-runner/main.py",
+ "entrypoint": "operator.py",
+ "metadata": "operator.json",
+ "name": "robomimic HDF5 示教校验",
+ "operatorId": "operator-builtin-robomimic-hdf5-validate",
+ "operatorSlug": "robomimic_hdf5_validate",
+ "preferredSlug": "robomimic_hdf5_validate",
+ "pythonImport": "backend.operators.builtin.robomimic_hdf5_validate.operator:run",
+ "repositoryId": "builtin/robomimic_hdf5_validate",
+ "runtimeDependencies": [
+ "robomimic-hdf5-files",
+ "h5py",
+ "numpy"
+ ],
+ "runtimeHandler": "builtin_robomimic_hdf5_validate",
+ "runtimeSource": "backend/operators/builtin/robomimic_hdf5_validate/operator.py",
+ "schemaVersion": "droboticflow.operator.repository/v1",
+ "sourceRoot": "backend/operators/builtin/robomimic_hdf5_validate",
+ "strictPocAcceptanceExternals": [],
+ "type": "builtin"
+}
diff --git a/backend/operators/builtin/tf_transform_tree_validate/__init__.py b/backend/operators/builtin/tf_transform_tree_validate/__init__.py
new file mode 100644
index 00000000..8b137891
--- /dev/null
+++ b/backend/operators/builtin/tf_transform_tree_validate/__init__.py
@@ -0,0 +1 @@
+
diff --git a/backend/operators/builtin/tf_transform_tree_validate/operator.json b/backend/operators/builtin/tf_transform_tree_validate/operator.json
new file mode 100644
index 00000000..4c44a749
--- /dev/null
+++ b/backend/operators/builtin/tf_transform_tree_validate/operator.json
@@ -0,0 +1,16 @@
+{
+ "id": "operator-builtin-tf-transform-tree-validate",
+ "name": "TF 坐标树校验",
+ "category": "ROBOTICS",
+ "type": "builtin",
+ "runtimeHandler": "builtin_tf_transform_tree_validate",
+ "runtimeSource": "backend/operators/builtin/tf_transform_tree_validate/operator.py",
+ "migrationStatus": "migrated",
+ "acceptance": {
+ "requiresRealInputs": true,
+ "mockOnlyAccepted": false,
+ "simulationOnlyAccepted": false,
+ "hardcodedOutputAccepted": false,
+ "shallowOnlyAccepted": false
+ }
+}
diff --git a/backend/operators/builtin/tf_transform_tree_validate/operator.py b/backend/operators/builtin/tf_transform_tree_validate/operator.py
new file mode 100644
index 00000000..a87e4e25
--- /dev/null
+++ b/backend/operators/builtin/tf_transform_tree_validate/operator.py
@@ -0,0 +1,294 @@
+from __future__ import annotations
+
+import csv
+import json
+import math
+import pathlib
+from typing import Any, Callable
+
+
+HelperMap = dict[str, Callable[..., Any]]
+TRANSFORM_SUFFIXES = {".json", ".jsonl", ".csv", ".yaml", ".yml"}
+
+
+def _int_param(params: dict[str, str], key: str, default: int) -> int:
+ raw = str(params.get(key, str(default)) or str(default)).strip()
+ try:
+ value = int(raw)
+ except ValueError as exc:
+ raise RuntimeError(f"{key} must be an integer") from exc
+ if value <= 0:
+ raise RuntimeError(f"{key} must be > 0")
+ return value
+
+
+def _float_param(params: dict[str, str], key: str, default: float) -> float:
+ raw = str(params.get(key, str(default)) or str(default)).strip()
+ try:
+ value = float(raw)
+ except ValueError as exc:
+ raise RuntimeError(f"{key} must be a number") from exc
+ if not math.isfinite(value) or value < 0:
+ raise RuntimeError(f"{key} must be a non-negative finite number")
+ return value
+
+
+def _find_transform_files(source_path: pathlib.Path, max_files: int) -> list[pathlib.Path]:
+ if source_path.is_file():
+ return [source_path] if source_path.suffix.lower() in TRANSFORM_SUFFIXES else []
+ preferred = [
+ source_path / "transforms.json",
+ source_path / "tf.json",
+ source_path / "tf_static.json",
+ source_path / "transforms.jsonl",
+ source_path / "transforms.csv",
+ source_path / "transforms.yaml",
+ ]
+ files = [item for item in preferred if item.is_file()]
+ files.extend(
+ item
+ for item in sorted(source_path.rglob("*"))
+ if item.is_file() and item.suffix.lower() in TRANSFORM_SUFFIXES and item not in files
+ )
+ return files[:max_files]
+
+
+def _load_yaml(path: pathlib.Path) -> Any:
+ try:
+ import yaml # type: ignore
+ except Exception as exc: # pragma: no cover - runtime dependency check
+ raise RuntimeError("TF 坐标树校验需要 PyYAML 以读取 YAML transform 文件") from exc
+ return yaml.safe_load(path.read_text(encoding="utf-8"))
+
+
+def _load_records(path: pathlib.Path) -> list[dict[str, Any]]:
+ suffix = path.suffix.lower()
+ if suffix == ".jsonl":
+ rows = [json.loads(line) for line in path.read_text(encoding="utf-8").splitlines() if line.strip()]
+ elif suffix == ".json":
+ payload = json.loads(path.read_text(encoding="utf-8"))
+ rows = payload.get("transforms", payload.get("items", payload)) if isinstance(payload, dict) else payload
+ elif suffix == ".csv":
+ with path.open("r", encoding="utf-8", newline="") as handle:
+ rows = list(csv.DictReader(handle))
+ else:
+ payload = _load_yaml(path)
+ rows = payload.get("transforms", payload.get("items", payload)) if isinstance(payload, dict) else payload
+ if isinstance(rows, dict):
+ rows = [rows]
+ if not isinstance(rows, list):
+ raise RuntimeError(f"{path} must contain a transform object or list")
+ return [row for row in rows if isinstance(row, dict)]
+
+
+def _nested(payload: dict[str, Any], dotted: str) -> Any:
+ current: Any = payload
+ for part in dotted.split("."):
+ if not isinstance(current, dict) or part not in current:
+ return None
+ current = current[part]
+ return current
+
+
+def _first_string(payload: dict[str, Any], keys: tuple[str, ...]) -> str:
+ for key in keys:
+ value = _nested(payload, key) if "." in key else payload.get(key)
+ if value is None:
+ continue
+ text = str(value).strip()
+ if text:
+ return text.lstrip("/")
+ return ""
+
+
+def _float_value(value: Any) -> float | None:
+ if isinstance(value, (int, float)):
+ number = float(value)
+ elif isinstance(value, str) and value.strip():
+ try:
+ number = float(value)
+ except ValueError:
+ return None
+ else:
+ return None
+ return number if math.isfinite(number) else None
+
+
+def _vector(payload: dict[str, Any], keys: tuple[str, str, str], nested_prefix: str = "") -> list[float] | None:
+ values: list[float] = []
+ for key in keys:
+ value = _nested(payload, f"{nested_prefix}.{key}") if nested_prefix else payload.get(key)
+ number = _float_value(value)
+ if number is None:
+ return None
+ values.append(number)
+ return values
+
+
+def _quaternion(payload: dict[str, Any]) -> list[float] | None:
+ candidates = [
+ ("transform.rotation", ("x", "y", "z", "w")),
+ ("rotation", ("x", "y", "z", "w")),
+ ("", ("qx", "qy", "qz", "qw")),
+ ]
+ for prefix, keys in candidates:
+ values: list[float] = []
+ for key in keys:
+ value = _nested(payload, f"{prefix}.{key}") if prefix else payload.get(key)
+ number = _float_value(value)
+ if number is None:
+ break
+ values.append(number)
+ if len(values) == 4:
+ return values
+ return None
+
+
+def _translation(payload: dict[str, Any]) -> list[float] | None:
+ for prefix in ("transform.translation", "translation", ""):
+ result = _vector(payload, ("x", "y", "z"), nested_prefix=prefix)
+ if result is not None:
+ return result
+ return None
+
+
+def _normalize_record(file_path: pathlib.Path, index: int, payload: dict[str, Any], tolerance: float) -> dict[str, Any]:
+ parent = _first_string(payload, ("header.frame_id", "parent_frame", "parent", "frame_id", "source_frame"))
+ child = _first_string(payload, ("child_frame_id", "child_frame", "child", "target_frame"))
+ translation = _translation(payload)
+ rotation = _quaternion(payload)
+ issues: list[dict[str, Any]] = []
+ if not parent:
+ issues.append({"code": "missing_parent_frame"})
+ if not child:
+ issues.append({"code": "missing_child_frame"})
+ if parent and child and parent == child:
+ issues.append({"code": "self_transform", "frame": parent})
+ if translation is None:
+ issues.append({"code": "missing_or_invalid_translation"})
+ if rotation is None:
+ issues.append({"code": "missing_or_invalid_quaternion"})
+ norm = 0.0
+ else:
+ norm = math.sqrt(sum(value * value for value in rotation))
+ if abs(norm - 1.0) > tolerance:
+ issues.append({"code": "quaternion_norm_out_of_tolerance", "norm": norm, "tolerance": tolerance})
+ return {
+ "file": str(file_path),
+ "index": index,
+ "valid": not issues,
+ "issueCount": len(issues),
+ "issues": issues,
+ "parentFrame": parent,
+ "childFrame": child,
+ "translation": translation or [],
+ "rotation": rotation or [],
+ "quaternionNorm": norm,
+ }
+
+
+def _cycle_edges(edges: dict[str, str]) -> list[list[str]]:
+ cycles: list[list[str]] = []
+ visited: set[str] = set()
+ for start in sorted(edges):
+ path: list[str] = []
+ seen_at: dict[str, int] = {}
+ node = start
+ while node in edges:
+ if node in seen_at:
+ cycle = path[seen_at[node] :]
+ if cycle and cycle not in cycles:
+ cycles.append(cycle)
+ break
+ if node in visited:
+ break
+ seen_at[node] = len(path)
+ path.append(node)
+ node = edges[node]
+ visited.update(path)
+ return cycles
+
+
+def run(
+ run_id: str,
+ node_id: str,
+ output_dir: Any,
+ node_data: dict[str, Any],
+ inputs: list[dict[str, Any]],
+ helpers: HelperMap,
+) -> list[dict[str, Any]]:
+ params = helpers["hyperparams"](node_data)
+ source_path = helpers["resolve_source_path"](params, inputs)
+ if source_path is None:
+ raise RuntimeError("TF 坐标树校验需要 SOURCE_PATH 指向真实 JSON/JSONL/CSV/YAML transform 文件")
+ max_files = _int_param(params, "MAX_FILES", 50)
+ max_transforms = _int_param(params, "MAX_TRANSFORMS", 10000)
+ tolerance = _float_param(params, "QUATERNION_NORM_TOLERANCE", 0.01)
+ files = _find_transform_files(source_path, max_files)
+ if not files:
+ raise RuntimeError(f"{source_path} does not contain transform files")
+
+ transforms: list[dict[str, Any]] = []
+ for file_path in files:
+ for row in _load_records(file_path):
+ if len(transforms) >= max_transforms:
+ break
+ transforms.append(_normalize_record(file_path, len(transforms), row, tolerance))
+
+ issues: list[dict[str, Any]] = []
+ issues.extend({"transformIndex": item["index"], **issue} for item in transforms for issue in item["issues"])
+ child_to_parent: dict[str, str] = {}
+ edge_counts: dict[tuple[str, str], int] = {}
+ frames: set[str] = set()
+ for item in transforms:
+ parent = str(item.get("parentFrame", ""))
+ child = str(item.get("childFrame", ""))
+ if not parent or not child or parent == child:
+ continue
+ frames.update([parent, child])
+ edge = (parent, child)
+ edge_counts[edge] = edge_counts.get(edge, 0) + 1
+ if child in child_to_parent and child_to_parent[child] != parent:
+ issues.append(
+ {
+ "code": "child_frame_has_multiple_parents",
+ "childFrame": child,
+ "parents": sorted({child_to_parent[child], parent}),
+ }
+ )
+ child_to_parent[child] = parent
+ for (parent, child), count in sorted(edge_counts.items()):
+ if count > 1:
+ issues.append({"code": "duplicate_transform_edge", "parentFrame": parent, "childFrame": child, "count": count})
+ cycles = _cycle_edges({child: parent for child, parent in child_to_parent.items()})
+ for cycle in cycles:
+ issues.append({"code": "cycle_detected", "frames": cycle})
+ child_frames = set(child_to_parent)
+ root_frames = sorted(frames - child_frames)
+ leaf_frames = sorted(frames - set(child_to_parent.values()))
+ payload = {
+ "datasetType": "tf-transform-tree-validation",
+ "sourcePath": str(source_path),
+ "valid": not issues,
+ "issueCount": len(issues),
+ "issues": issues[:300],
+ "fileCount": len(files),
+ "transformCount": len(transforms),
+ "validTransformCount": sum(1 for item in transforms if item["valid"]),
+ "frameCount": len(frames),
+ "rootFrames": root_frames,
+ "rootFrameCount": len(root_frames),
+ "leafFrames": leaf_frames,
+ "cycleCount": len(cycles),
+ "transforms": transforms[:200],
+ }
+ return [
+ helpers["write_json_artifact"](
+ run_id,
+ node_id,
+ output_dir,
+ "tf-transform-tree-validation.json",
+ payload,
+ output=helpers["output_binding"](node_data),
+ )
+ ]
diff --git a/backend/operators/builtin/tf_transform_tree_validate/repository.json b/backend/operators/builtin/tf_transform_tree_validate/repository.json
new file mode 100644
index 00000000..68b311a3
--- /dev/null
+++ b/backend/operators/builtin/tf_transform_tree_validate/repository.json
@@ -0,0 +1,31 @@
+{
+ "acceptance": {
+ "hardcodedOutputAccepted": false,
+ "mockOnlyAccepted": false,
+ "pocFullAcceptanceScript": "scripts/poc-operator-full-acceptance.sh",
+ "requiresPersistentWorkspaceEvidence": true,
+ "requiresRealInputs": true,
+ "shallowOnlyAccepted": false,
+ "simulationOnlyAccepted": false
+ },
+ "category": "ROBOTICS",
+ "containerEntrypoint": "backend/operator-runner/main.py",
+ "entrypoint": "operator.py",
+ "metadata": "operator.json",
+ "name": "TF 坐标树校验",
+ "operatorId": "operator-builtin-tf-transform-tree-validate",
+ "operatorSlug": "tf_transform_tree_validate",
+ "preferredSlug": "tf_transform_tree_validate",
+ "pythonImport": "backend.operators.builtin.tf_transform_tree_validate.operator:run",
+ "repositoryId": "builtin/tf_transform_tree_validate",
+ "runtimeDependencies": [
+ "json-jsonl-csv-yaml-files",
+ "optional-pyyaml"
+ ],
+ "runtimeHandler": "builtin_tf_transform_tree_validate",
+ "runtimeSource": "backend/operators/builtin/tf_transform_tree_validate/operator.py",
+ "schemaVersion": "droboticflow.operator.repository/v1",
+ "sourceRoot": "backend/operators/builtin/tf_transform_tree_validate",
+ "strictPocAcceptanceExternals": [],
+ "type": "builtin"
+}
diff --git a/docs/operators/MJCF-执行器校验.md b/docs/operators/MJCF-执行器校验.md
new file mode 100644
index 00000000..ba9fa117
--- /dev/null
+++ b/docs/operators/MJCF-执行器校验.md
@@ -0,0 +1,109 @@
+# MJCF 执行器校验
+
+- 算子 ID:`operator-builtin-mjcf-actuator-validate`
+- 类别:`ROBOTICS`
+- 类型:`builtin`
+
+## 作用
+
+读取真实 MuJoCo/MJCF XML,校验 actuator 下 motor/position/velocity/general 等执行器目标引用、ctrlrange/forcerange/actrange 和 gear 数值。
+
+## 定位与职责边界
+
+- 该算子属于 `ROBOTICS` 类,当前登记类型是 `builtin`。
+- 平台会在 `/api/operators` 中自动提供该算子元数据,运行时由 operator-runner 委派到独立算子入口。
+
+## 平台语义
+
+面向 MuJoCo、robosuite、ManiSkill 和 GitHub 常见 MJCF 仿真资产,真实解析 下执行器,验证 joint/tendon/site/body 引用存在,ctrlrange/forcerange/actrange 为递增有限数值,并识别重复 actuator 名称。
+
+## 运行时契约
+
+- Worker 会先把上游 artifact 汇总到 `INPUT_ARTIFACTS_PATH`,再把节点配置通过 `NODE_DATA_JSON` 注入运行容器。
+- 节点执行结果会写入 `OUTPUT_DIR`、`RESULT_OUTPUT_PATH` 和 `STAGE_MANIFEST_PATH` 对应的文件,再由 worker 汇总进运行记录。
+- 大文件和大 JSON 不应依赖 API 内联传输,应该通过 `artifact.path`、`dataRef.payloadPath`、共享 PVC 或对象存储引用传递。
+- 当前独立实现入口是 `backend/operators/builtin/mjcf_actuator_validate/operator.py` 中的 `run(...)`,平台 runtimeHandler 为 `builtin_mjcf_actuator_validate`。
+
+## 设计边界
+
+- 该算子应该只承担单一职责,不应在一个节点里同时混合索引、筛选、上传、审核等多个阶段性行为。
+- 如果一个节点要消费多个语义不同的结果,优先通过多节点拆分,而不是把所有逻辑塞进一个超大脚本。
+
+## 数据流示意
+
+```mermaid
+flowchart LR
+ A[上游输入或运行时注入] --> B[MJCF 执行器校验]
+ B --> C[artifact / stage manifest / 外部副作用]
+```
+
+## 输入端口详解
+
+- `in-1` / **input** / `json` / `bus`:MJCF XML 文件或包含 MJCF 文件的目录 作为输入端口时,适合结构化清单、规则结果、摘要对象等可直接序列化的数据。 通过统一 artifact 契约传递;大 payload 会外置到 stage manifest / payloadPath。
+
+## 输出端口详解
+
+- `out-1` / **validation** / `json` / `bus`:MJCF 执行器校验结果 作为输出端口时,适合结构化清单、规则结果、摘要对象等可直接序列化的数据。 通过统一 artifact 契约传递;大 payload 会外置到 stage manifest / payloadPath。
+
+## 关键超参数
+
+- `SOURCE_PATH`:类型 `string`,默认值 空字符串或由工作流显式填写。MJCF XML 文件或包含 MJCF 文件的目录 通常建议在工作流中显式配置。
+- `MAX_MODELS`:类型 `string`,默认值 `20`。最多扫描多少个 MJCF 文件,必须为正整数 可直接使用默认值,也可在节点级覆盖。
+
+## 执行前准备
+
+- 先确认上下游端口类型兼容,特别是 `json`、`file`、`video` 等端口不要混接。
+- 为每个关键超参数补齐明确值,避免依赖隐式默认值进入生产工作流。
+- 对需要跨节点长期保留的产物,提前规划对象存储前缀或外部落库集合,而不是只写临时目录。
+
+## 数据通信与存储路径
+
+- 标准输入来自 `INPUT_ARTIFACTS_PATH` 指向的 artifact 列表文件,artifact 可同时包含 `data`、`path`、`metadata` 与 `dataRef`。
+- 标准输出写入 `OUTPUT_DIR`,并通过 `stage-manifest.json` 对外暴露产物路径、payload 引用和摘要元数据。
+
+## 典型上下游编排
+
+- 该算子通常位于同类数据处理节点之间,承担结构化中间层职责。
+
+## 节点配置示例
+
+```json
+{
+ "label": "MJCF 执行器校验",
+ "operatorId": "operator-builtin-mjcf-actuator-validate",
+ "type": "builtin",
+ "category": "ROBOTICS",
+ "hyperparams": {
+ "SOURCE_PATH": "<按节点填写>",
+ "MAX_MODELS": "20"
+ }
+}
+```
+
+## 输入输出示例
+
+- 典型输入:worker 会把上游 artifact 序列化为 `INPUT_ARTIFACTS_PATH` 文件供节点读取。
+- 典型输出:节点会在 `OUTPUT_DIR` 下写业务文件,并在 `stage-manifest.json` 里登记对外可消费的结果。
+
+## 成功判定
+
+- 节点 execution 状态为 `completed`,且日志中没有隐藏的回退错误或重试未果记录。
+- `stage-manifest.json` 中能看到期望的 artifact 条目,且下游节点可以正常消费这些结果。
+
+## 常见失败与排查
+
+- 如果看不到预期结果,先检查 `OUTPUT_DIR`、`RESULT_OUTPUT_PATH` 和 `STAGE_MANIFEST_PATH` 是否真实写出。
+- 如果下游收不到数据,先检查当前节点输出 artifact 是否设置了正确的 `portId`、`portName` 和 `type`。
+- 如果日志中只有摘要没有大 payload,优先排查 `dataRef.payloadPath` 或外置文件是否存在,而不是误以为结果丢失。
+
+## 验收检查清单
+
+- 至少在一条真实工作流中跑通该算子,而不是只验证单元测试或空输入。
+- 确认该节点输出被下游节点实际消费,避免出现“本节点成功但链路无效”的假阳性。
+- 保留一次成功运行的示例日志、stage manifest 和关键产物路径,便于后续回归对照。
+
+## 相关文档
+
+- [ROBOTICS 类索引](README.md)
+- [算子开发文档](../operator-authoring.md)
+- [内建算子总手册](../builtin-operators-manual.zh-CN.md)
diff --git a/docs/operators/README.md b/docs/operators/README.md
index 14d56e52..206e0065 100644
--- a/docs/operators/README.md
+++ b/docs/operators/README.md
@@ -2,7 +2,7 @@
当前目录按“每个内建算子一份 Markdown”组织,内容基于当前平台真实算子定义按算子名称去重生成。
-- 当前收录数量:`55`
+- 当前收录数量:`58`
- 文档定位:单算子详细手册,重点覆盖输入输出契约、运行时路径、配置示例、排障与验收。
- 配套总览:[`../builtin-operators-manual.zh-CN.md`](../builtin-operators-manual.zh-CN.md)
@@ -72,13 +72,16 @@
- [LeRobot 数据集校验](LeRobot-数据集校验.md)
- [LeRobot 数据集概览](LeRobot-数据集概览.md)
- [MCAP 主题概览](MCAP-主题概览.md)
+- [MJCF 执行器校验](MJCF-执行器校验.md)
- [RGB-D 帧对齐校验](RGB-D-帧对齐校验.md)
- [RLDS 轨迹概览](RLDS-轨迹概览.md)
- [ROSBag 数据概览](ROSBag-数据概览.md)
+- [TF 坐标树校验](TF-坐标树校验.md)
- [TFRecord 结构校验](TFRecord-结构校验.md)
- [URDF 关节限位校验](URDF-关节限位校验.md)
- [VLA 样本清单校验](VLA-样本清单校验.md)
- [Zarr 轨迹数组概览](Zarr-轨迹数组概览.md)
+- [robomimic HDF5 示教校验](robomimic-HDF5-示教校验.md)
- [分割 Mask 校验](分割-Mask-校验.md)
- [动作状态数值质量校验](动作状态数值质量校验.md)
- [多传感器时间同步校验](多传感器时间同步校验.md)
diff --git a/docs/operators/TF-坐标树校验.md b/docs/operators/TF-坐标树校验.md
new file mode 100644
index 00000000..c79cf6ac
--- /dev/null
+++ b/docs/operators/TF-坐标树校验.md
@@ -0,0 +1,113 @@
+# TF 坐标树校验
+
+- 算子 ID:`operator-builtin-tf-transform-tree-validate`
+- 类别:`ROBOTICS`
+- 类型:`builtin`
+
+## 作用
+
+读取真实 ROS TransformStamped 风格 JSON/JSONL/YAML 或扁平 CSV 坐标变换,校验父子 frame、平移、四元数归一化、多父节点、重复边和环。
+
+## 定位与职责边界
+
+- 该算子属于 `ROBOTICS` 类,当前登记类型是 `builtin`。
+- 平台会在 `/api/operators` 中自动提供该算子元数据,运行时由 operator-runner 委派到独立算子入口。
+
+## 平台语义
+
+面向 ROS/ROS2 机器人数据集、相机外参和多传感器标定入库,真实读取 header.frame_id、child_frame_id、translation、rotation 或 parent_frame/child_frame/qx/qy/qz/qw,校验坐标树根、叶子、重复边、多父节点和环路。
+
+## 运行时契约
+
+- Worker 会先把上游 artifact 汇总到 `INPUT_ARTIFACTS_PATH`,再把节点配置通过 `NODE_DATA_JSON` 注入运行容器。
+- 节点执行结果会写入 `OUTPUT_DIR`、`RESULT_OUTPUT_PATH` 和 `STAGE_MANIFEST_PATH` 对应的文件,再由 worker 汇总进运行记录。
+- 大文件和大 JSON 不应依赖 API 内联传输,应该通过 `artifact.path`、`dataRef.payloadPath`、共享 PVC 或对象存储引用传递。
+- 当前独立实现入口是 `backend/operators/builtin/tf_transform_tree_validate/operator.py` 中的 `run(...)`,平台 runtimeHandler 为 `builtin_tf_transform_tree_validate`。
+
+## 设计边界
+
+- 该算子应该只承担单一职责,不应在一个节点里同时混合索引、筛选、上传、审核等多个阶段性行为。
+- 如果一个节点要消费多个语义不同的结果,优先通过多节点拆分,而不是把所有逻辑塞进一个超大脚本。
+
+## 数据流示意
+
+```mermaid
+flowchart LR
+ A[上游输入或运行时注入] --> B[TF 坐标树校验]
+ B --> C[artifact / stage manifest / 外部副作用]
+```
+
+## 输入端口详解
+
+- `in-1` / **input** / `json` / `bus`:transforms/tf/tf_static JSON、JSONL、CSV 或 YAML 文件/目录 作为输入端口时,适合结构化清单、规则结果、摘要对象等可直接序列化的数据。 通过统一 artifact 契约传递;大 payload 会外置到 stage manifest / payloadPath。
+
+## 输出端口详解
+
+- `out-1` / **validation** / `json` / `bus`:TF 坐标树校验结果 作为输出端口时,适合结构化清单、规则结果、摘要对象等可直接序列化的数据。 通过统一 artifact 契约传递;大 payload 会外置到 stage manifest / payloadPath。
+
+## 关键超参数
+
+- `SOURCE_PATH`:类型 `string`,默认值 空字符串或由工作流显式填写。坐标变换文件或包含坐标变换文件的目录 通常建议在工作流中显式配置。
+- `QUATERNION_NORM_TOLERANCE`:类型 `string`,默认值 `0.01`。四元数 norm 与 1 的允许误差 可直接使用默认值,也可在节点级覆盖。
+- `MAX_TRANSFORMS`:类型 `string`,默认值 `10000`。最多校验多少条 transform 记录,必须为正整数 可直接使用默认值,也可在节点级覆盖。
+- `MAX_FILES`:类型 `string`,默认值 `50`。最多扫描多少个 transform 文件,必须为正整数 可直接使用默认值,也可在节点级覆盖。
+
+## 执行前准备
+
+- 先确认上下游端口类型兼容,特别是 `json`、`file`、`video` 等端口不要混接。
+- 为每个关键超参数补齐明确值,避免依赖隐式默认值进入生产工作流。
+- 对需要跨节点长期保留的产物,提前规划对象存储前缀或外部落库集合,而不是只写临时目录。
+
+## 数据通信与存储路径
+
+- 标准输入来自 `INPUT_ARTIFACTS_PATH` 指向的 artifact 列表文件,artifact 可同时包含 `data`、`path`、`metadata` 与 `dataRef`。
+- 标准输出写入 `OUTPUT_DIR`,并通过 `stage-manifest.json` 对外暴露产物路径、payload 引用和摘要元数据。
+
+## 典型上下游编排
+
+- 该算子通常位于同类数据处理节点之间,承担结构化中间层职责。
+
+## 节点配置示例
+
+```json
+{
+ "label": "TF 坐标树校验",
+ "operatorId": "operator-builtin-tf-transform-tree-validate",
+ "type": "builtin",
+ "category": "ROBOTICS",
+ "hyperparams": {
+ "SOURCE_PATH": "<按节点填写>",
+ "QUATERNION_NORM_TOLERANCE": "0.01",
+ "MAX_TRANSFORMS": "10000",
+ "MAX_FILES": "50"
+ }
+}
+```
+
+## 输入输出示例
+
+- 典型输入:worker 会把上游 artifact 序列化为 `INPUT_ARTIFACTS_PATH` 文件供节点读取。
+- 典型输出:节点会在 `OUTPUT_DIR` 下写业务文件,并在 `stage-manifest.json` 里登记对外可消费的结果。
+
+## 成功判定
+
+- 节点 execution 状态为 `completed`,且日志中没有隐藏的回退错误或重试未果记录。
+- `stage-manifest.json` 中能看到期望的 artifact 条目,且下游节点可以正常消费这些结果。
+
+## 常见失败与排查
+
+- 如果看不到预期结果,先检查 `OUTPUT_DIR`、`RESULT_OUTPUT_PATH` 和 `STAGE_MANIFEST_PATH` 是否真实写出。
+- 如果下游收不到数据,先检查当前节点输出 artifact 是否设置了正确的 `portId`、`portName` 和 `type`。
+- 如果日志中只有摘要没有大 payload,优先排查 `dataRef.payloadPath` 或外置文件是否存在,而不是误以为结果丢失。
+
+## 验收检查清单
+
+- 至少在一条真实工作流中跑通该算子,而不是只验证单元测试或空输入。
+- 确认该节点输出被下游节点实际消费,避免出现“本节点成功但链路无效”的假阳性。
+- 保留一次成功运行的示例日志、stage manifest 和关键产物路径,便于后续回归对照。
+
+## 相关文档
+
+- [ROBOTICS 类索引](README.md)
+- [算子开发文档](../operator-authoring.md)
+- [内建算子总手册](../builtin-operators-manual.zh-CN.md)
diff --git a/docs/operators/robomimic-HDF5-示教校验.md b/docs/operators/robomimic-HDF5-示教校验.md
new file mode 100644
index 00000000..640d1a6b
--- /dev/null
+++ b/docs/operators/robomimic-HDF5-示教校验.md
@@ -0,0 +1,121 @@
+# robomimic HDF5 示教校验
+
+- 算子 ID:`operator-builtin-robomimic-hdf5-validate`
+- 类别:`ROBOTICS`
+- 类型:`builtin`
+
+## 作用
+
+读取真实 robomimic HDF5 示教文件,校验 /data/demo_* 下 actions、rewards、dones、obs、next_obs、states、mask 与 total 样本数一致性。
+
+## 定位与职责边界
+
+- 该算子属于 `ROBOTICS` 类,当前登记类型是 `builtin`。
+- 平台会在 `/api/operators` 中自动提供该算子元数据,运行时由 operator-runner 委派到独立算子入口。
+
+## 平台语义
+
+面向 robomimic、D4RL/robosuite 派生示教数据和 GitHub 常见机器人模仿学习 HDF5 数据,真实读取 /data/demo_* group,校验 actions/rewards/dones 与 obs/next_obs 帧数、num_samples、data.attrs.total 和 mask/train/valid 引用关系。
+
+## 运行时契约
+
+- Worker 会先把上游 artifact 汇总到 `INPUT_ARTIFACTS_PATH`,再把节点配置通过 `NODE_DATA_JSON` 注入运行容器。
+- 节点执行结果会写入 `OUTPUT_DIR`、`RESULT_OUTPUT_PATH` 和 `STAGE_MANIFEST_PATH` 对应的文件,再由 worker 汇总进运行记录。
+- 大文件和大 JSON 不应依赖 API 内联传输,应该通过 `artifact.path`、`dataRef.payloadPath`、共享 PVC 或对象存储引用传递。
+- 当前独立实现入口是 `backend/operators/builtin/robomimic_hdf5_validate/operator.py` 中的 `run(...)`,平台 runtimeHandler 为 `builtin_robomimic_hdf5_validate`。
+
+## 设计边界
+
+- 该算子应该只承担单一职责,不应在一个节点里同时混合索引、筛选、上传、审核等多个阶段性行为。
+- 如果一个节点要消费多个语义不同的结果,优先通过多节点拆分,而不是把所有逻辑塞进一个超大脚本。
+
+## 数据流示意
+
+```mermaid
+flowchart LR
+ A[上游输入或运行时注入] --> B[robomimic HDF5 示教校验]
+ B --> C[artifact / stage manifest / 外部副作用]
+```
+
+## 输入端口详解
+
+- `in-1` / **input** / `json` / `bus`:robomimic .hdf5/.h5 文件或目录输入 作为输入端口时,适合结构化清单、规则结果、摘要对象等可直接序列化的数据。 通过统一 artifact 契约传递;大 payload 会外置到 stage manifest / payloadPath。
+
+## 输出端口详解
+
+- `out-1` / **validation** / `json` / `bus`:robomimic HDF5 示教结构校验结果 作为输出端口时,适合结构化清单、规则结果、摘要对象等可直接序列化的数据。 通过统一 artifact 契约传递;大 payload 会外置到 stage manifest / payloadPath。
+
+## 关键超参数
+
+- `SOURCE_PATH`:类型 `string`,默认值 空字符串或由工作流显式填写。.hdf5/.h5 文件或包含 robomimic HDF5 文件的目录 通常建议在工作流中显式配置。
+- `REQUIRE_REWARDS`:类型 `string`,默认值 `true`。是否要求每个 demo 包含 rewards 数据集 可直接使用默认值,也可在节点级覆盖。
+- `REQUIRE_DONES`:类型 `string`,默认值 `true`。是否要求每个 demo 包含 dones 数据集 可直接使用默认值,也可在节点级覆盖。
+- `REQUIRE_NEXT_OBS`:类型 `string`,默认值 `true`。是否要求每个 demo 包含 next_obs group 可直接使用默认值,也可在节点级覆盖。
+- `CHECK_ACTION_RANGE`:类型 `string`,默认值 `false`。是否检查 action 绝对值不超过 MAX_ABS_ACTION;默认关闭以兼容未归一化控制量 可直接使用默认值,也可在节点级覆盖。
+- `MAX_ABS_ACTION`:类型 `string`,默认值 `1.0`。开启 CHECK_ACTION_RANGE 后允许的最大 action 绝对值 可直接使用默认值,也可在节点级覆盖。
+- `MAX_DEMOS`:类型 `string`,默认值 `200`。每个文件最多校验多少个 demo,必须为正整数 可直接使用默认值,也可在节点级覆盖。
+- `MAX_FILES`:类型 `string`,默认值 `20`。最多校验多少个 HDF5 文件,必须为正整数 可直接使用默认值,也可在节点级覆盖。
+
+## 执行前准备
+
+- 先确认上下游端口类型兼容,特别是 `json`、`file`、`video` 等端口不要混接。
+- 为每个关键超参数补齐明确值,避免依赖隐式默认值进入生产工作流。
+- 对需要跨节点长期保留的产物,提前规划对象存储前缀或外部落库集合,而不是只写临时目录。
+
+## 数据通信与存储路径
+
+- 标准输入来自 `INPUT_ARTIFACTS_PATH` 指向的 artifact 列表文件,artifact 可同时包含 `data`、`path`、`metadata` 与 `dataRef`。
+- 标准输出写入 `OUTPUT_DIR`,并通过 `stage-manifest.json` 对外暴露产物路径、payload 引用和摘要元数据。
+
+## 典型上下游编排
+
+- 该算子通常位于同类数据处理节点之间,承担结构化中间层职责。
+
+## 节点配置示例
+
+```json
+{
+ "label": "robomimic HDF5 示教校验",
+ "operatorId": "operator-builtin-robomimic-hdf5-validate",
+ "type": "builtin",
+ "category": "ROBOTICS",
+ "hyperparams": {
+ "SOURCE_PATH": "<按节点填写>",
+ "REQUIRE_REWARDS": "true",
+ "REQUIRE_DONES": "true",
+ "REQUIRE_NEXT_OBS": "true",
+ "CHECK_ACTION_RANGE": "false",
+ "MAX_ABS_ACTION": "1.0",
+ "MAX_DEMOS": "200",
+ "MAX_FILES": "20"
+ }
+}
+```
+
+## 输入输出示例
+
+- 典型输入:worker 会把上游 artifact 序列化为 `INPUT_ARTIFACTS_PATH` 文件供节点读取。
+- 典型输出:节点会在 `OUTPUT_DIR` 下写业务文件,并在 `stage-manifest.json` 里登记对外可消费的结果。
+
+## 成功判定
+
+- 节点 execution 状态为 `completed`,且日志中没有隐藏的回退错误或重试未果记录。
+- `stage-manifest.json` 中能看到期望的 artifact 条目,且下游节点可以正常消费这些结果。
+
+## 常见失败与排查
+
+- 如果看不到预期结果,先检查 `OUTPUT_DIR`、`RESULT_OUTPUT_PATH` 和 `STAGE_MANIFEST_PATH` 是否真实写出。
+- 如果下游收不到数据,先检查当前节点输出 artifact 是否设置了正确的 `portId`、`portName` 和 `type`。
+- 如果日志中只有摘要没有大 payload,优先排查 `dataRef.payloadPath` 或外置文件是否存在,而不是误以为结果丢失。
+
+## 验收检查清单
+
+- 至少在一条真实工作流中跑通该算子,而不是只验证单元测试或空输入。
+- 确认该节点输出被下游节点实际消费,避免出现“本节点成功但链路无效”的假阳性。
+- 保留一次成功运行的示例日志、stage manifest 和关键产物路径,便于后续回归对照。
+
+## 相关文档
+
+- [ROBOTICS 类索引](README.md)
+- [算子开发文档](../operator-authoring.md)
+- [内建算子总手册](../builtin-operators-manual.zh-CN.md)
diff --git a/internal/api/builtin_operators_catalog.json b/internal/api/builtin_operators_catalog.json
index 77bbc146..ddf0486f 100644
--- a/internal/api/builtin_operators_catalog.json
+++ b/internal/api/builtin_operators_catalog.json
@@ -1490,6 +1490,84 @@
"runtimeSource": "backend/operators/builtin/hdf5_demonstration_summary/operator.py",
"runtimeHandler": "builtin_hdf5_demonstration_summary"
},
+ {
+ "id": "operator-builtin-robomimic-hdf5-validate",
+ "name": "robomimic HDF5 示教校验",
+ "type": "builtin",
+ "category": "ROBOTICS",
+ "description": "读取真实 robomimic HDF5 示教文件,校验 /data/demo_* 下 actions、rewards、dones、obs、next_obs、states、mask 与 total 样本数一致性。",
+ "inputs": [
+ {
+ "id": "in-1",
+ "name": "input",
+ "type": "json",
+ "description": "robomimic .hdf5/.h5 文件或目录输入",
+ "transmission": "bus"
+ }
+ ],
+ "outputs": [
+ {
+ "id": "out-1",
+ "name": "validation",
+ "type": "json",
+ "description": "robomimic HDF5 示教结构校验结果",
+ "transmission": "bus"
+ }
+ ],
+ "skillDefinition": "面向 robomimic、D4RL/robosuite 派生示教数据和 GitHub 常见机器人模仿学习 HDF5 数据,真实读取 /data/demo_* group,校验 actions/rewards/dones 与 obs/next_obs 帧数、num_samples、data.attrs.total 和 mask/train/valid 引用关系。",
+ "hyperparams": [
+ {
+ "key": "SOURCE_PATH",
+ "value": "",
+ "description": ".hdf5/.h5 文件或包含 robomimic HDF5 文件的目录",
+ "type": "string"
+ },
+ {
+ "key": "REQUIRE_REWARDS",
+ "value": "true",
+ "description": "是否要求每个 demo 包含 rewards 数据集",
+ "type": "string"
+ },
+ {
+ "key": "REQUIRE_DONES",
+ "value": "true",
+ "description": "是否要求每个 demo 包含 dones 数据集",
+ "type": "string"
+ },
+ {
+ "key": "REQUIRE_NEXT_OBS",
+ "value": "true",
+ "description": "是否要求每个 demo 包含 next_obs group",
+ "type": "string"
+ },
+ {
+ "key": "CHECK_ACTION_RANGE",
+ "value": "false",
+ "description": "是否检查 action 绝对值不超过 MAX_ABS_ACTION;默认关闭以兼容未归一化控制量",
+ "type": "string"
+ },
+ {
+ "key": "MAX_ABS_ACTION",
+ "value": "1.0",
+ "description": "开启 CHECK_ACTION_RANGE 后允许的最大 action 绝对值",
+ "type": "string"
+ },
+ {
+ "key": "MAX_DEMOS",
+ "value": "200",
+ "description": "每个文件最多校验多少个 demo,必须为正整数",
+ "type": "string"
+ },
+ {
+ "key": "MAX_FILES",
+ "value": "20",
+ "description": "最多校验多少个 HDF5 文件,必须为正整数",
+ "type": "string"
+ }
+ ],
+ "runtimeSource": "backend/operators/builtin/robomimic_hdf5_validate/operator.py",
+ "runtimeHandler": "builtin_robomimic_hdf5_validate"
+ },
{
"id": "operator-builtin-aloha-hdf5-validate",
"name": "ALOHA HDF5 示教校验",
@@ -2114,6 +2192,102 @@
"runtimeSource": "backend/operators/builtin/camera_calibration_validate/operator.py",
"runtimeHandler": "builtin_camera_calibration_validate"
},
+ {
+ "id": "operator-builtin-tf-transform-tree-validate",
+ "name": "TF 坐标树校验",
+ "type": "builtin",
+ "category": "ROBOTICS",
+ "description": "读取真实 ROS TransformStamped 风格 JSON/JSONL/YAML 或扁平 CSV 坐标变换,校验父子 frame、平移、四元数归一化、多父节点、重复边和环。",
+ "inputs": [
+ {
+ "id": "in-1",
+ "name": "input",
+ "type": "json",
+ "description": "transforms/tf/tf_static JSON、JSONL、CSV 或 YAML 文件/目录",
+ "transmission": "bus"
+ }
+ ],
+ "outputs": [
+ {
+ "id": "out-1",
+ "name": "validation",
+ "type": "json",
+ "description": "TF 坐标树校验结果",
+ "transmission": "bus"
+ }
+ ],
+ "skillDefinition": "面向 ROS/ROS2 机器人数据集、相机外参和多传感器标定入库,真实读取 header.frame_id、child_frame_id、translation、rotation 或 parent_frame/child_frame/qx/qy/qz/qw,校验坐标树根、叶子、重复边、多父节点和环路。",
+ "hyperparams": [
+ {
+ "key": "SOURCE_PATH",
+ "value": "",
+ "description": "坐标变换文件或包含坐标变换文件的目录",
+ "type": "string"
+ },
+ {
+ "key": "QUATERNION_NORM_TOLERANCE",
+ "value": "0.01",
+ "description": "四元数 norm 与 1 的允许误差",
+ "type": "string"
+ },
+ {
+ "key": "MAX_TRANSFORMS",
+ "value": "10000",
+ "description": "最多校验多少条 transform 记录,必须为正整数",
+ "type": "string"
+ },
+ {
+ "key": "MAX_FILES",
+ "value": "50",
+ "description": "最多扫描多少个 transform 文件,必须为正整数",
+ "type": "string"
+ }
+ ],
+ "runtimeSource": "backend/operators/builtin/tf_transform_tree_validate/operator.py",
+ "runtimeHandler": "builtin_tf_transform_tree_validate"
+ },
+ {
+ "id": "operator-builtin-mjcf-actuator-validate",
+ "name": "MJCF 执行器校验",
+ "type": "builtin",
+ "category": "ROBOTICS",
+ "description": "读取真实 MuJoCo/MJCF XML,校验 actuator 下 motor/position/velocity/general 等执行器目标引用、ctrlrange/forcerange/actrange 和 gear 数值。",
+ "inputs": [
+ {
+ "id": "in-1",
+ "name": "input",
+ "type": "json",
+ "description": "MJCF XML 文件或包含 MJCF 文件的目录",
+ "transmission": "bus"
+ }
+ ],
+ "outputs": [
+ {
+ "id": "out-1",
+ "name": "validation",
+ "type": "json",
+ "description": "MJCF 执行器校验结果",
+ "transmission": "bus"
+ }
+ ],
+ "skillDefinition": "面向 MuJoCo、robosuite、ManiSkill 和 GitHub 常见 MJCF 仿真资产,真实解析 下执行器,验证 joint/tendon/site/body 引用存在,ctrlrange/forcerange/actrange 为递增有限数值,并识别重复 actuator 名称。",
+ "hyperparams": [
+ {
+ "key": "SOURCE_PATH",
+ "value": "",
+ "description": "MJCF XML 文件或包含 MJCF 文件的目录",
+ "type": "string"
+ },
+ {
+ "key": "MAX_MODELS",
+ "value": "20",
+ "description": "最多扫描多少个 MJCF 文件,必须为正整数",
+ "type": "string"
+ }
+ ],
+ "runtimeSource": "backend/operators/builtin/mjcf_actuator_validate/operator.py",
+ "runtimeHandler": "builtin_mjcf_actuator_validate"
+ },
{
"id": "operator-builtin-robot-model-asset-validate",
"name": "机器人模型资产校验",
diff --git a/scripts/generate_operator_repository_manifests.py b/scripts/generate_operator_repository_manifests.py
index cd24eb38..c29c768c 100644
--- a/scripts/generate_operator_repository_manifests.py
+++ b/scripts/generate_operator_repository_manifests.py
@@ -41,6 +41,7 @@ RUNTIME_DEPENDENCIES: dict[str, list[str]] = {
"lerobot_summary": ["lerobot-dataset-files"],
"mcap_topic_summary": ["mcap-files", "mcap-python"],
"merge": ["json-artifacts"],
+ "mjcf_actuator_validate": ["mjcf-files", "xml-parser"],
"model_inference": ["json-model-or-command"],
"mongodb_write": ["mongodb"],
"package": ["filesystem"],
@@ -49,6 +50,7 @@ RUNTIME_DEPENDENCIES: dict[str, list[str]] = {
"real_source": ["filesystem"],
"rgbd_frame_pair_validate": ["rgb-images", "depth-images", "opencv-python-headless", "numpy", "pyyaml"],
"rlds_episode_summary": ["rlds-json-or-tfds-files"],
+ "robomimic_hdf5_validate": ["robomimic-hdf5-files", "h5py", "numpy"],
"robot_model_asset_validate": ["urdf-files", "mjcf-files", "mesh-assets"],
"rosbag_summary": ["rosbag2-files", "sqlite"],
"sensor_timestamp_sync_validate": ["timestamp-csv-jsonl"],
@@ -56,6 +58,7 @@ RUNTIME_DEPENDENCIES: dict[str, list[str]] = {
"shell_command": ["container-shell"],
"template_render": ["json-artifacts"],
"tfrecord_structural_validate": ["tfrecord-files", "crc32c-structural-validation"],
+ "tf_transform_tree_validate": ["json-jsonl-csv-yaml-files", "optional-pyyaml"],
"trajectory_integrity_validate": ["trajectory-json", "local-media-files"],
"upload": ["filesystem", "optional-baidu-bos"],
"video_annotate": ["video-files", "json-artifacts"],