diff --git a/backend/operator-runner/main.py b/backend/operator-runner/main.py index 9a7d732d..faabd222 100644 --- a/backend/operator-runner/main.py +++ b/backend/operator-runner/main.py @@ -80,6 +80,7 @@ from backend.operators.builtin.pointcloud_summary import operator as pointcloud_ 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.rh20t_scene_validate import operator as rh20t_scene_validate_operator +from backend.operators.builtin.robomind_hdf5_quality_validate import operator as robomind_hdf5_quality_validate_operator from backend.operators.builtin.robomimic_hdf5_validate import operator as robomimic_hdf5_validate_operator from backend.operators.builtin.robocasa_hdf5_demonstration_validate import operator as robocasa_hdf5_demonstration_validate_operator from backend.operators.builtin.robot_model_asset_validate import operator as robot_model_asset_validate_operator @@ -4971,6 +4972,7 @@ BUILTIN_HANDLER_RULES: tuple[tuple[str, tuple[str, ...]], ...] = ( ("libero_hdf5_task_validate", ("libero hdf5", "libero 任务", "libero 数据集", "libero benchmark")), ("robocasa_hdf5_demonstration_validate", ("robocasa hdf5", "robocasa 示教", "robocasa 数据集", "robocasa demonstration")), ("rh20t_scene_validate", ("rh20t", "rh20t 场景", "rh20t scene", "real human 20t")), + ("robomind_hdf5_quality_validate", ("robomind hdf5", "robomind 质量", "robomind 数据", "multi-embodiment intelligence")), ("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")), @@ -5067,6 +5069,7 @@ BUILTIN_HANDLERS: dict[str, BuiltinHandler] = { "robomimic_hdf5_validate": operator_handler(robomimic_hdf5_validate_operator), "robocasa_hdf5_demonstration_validate": operator_handler(robocasa_hdf5_demonstration_validate_operator), "rh20t_scene_validate": operator_handler(rh20t_scene_validate_operator), + "robomind_hdf5_quality_validate": operator_handler(robomind_hdf5_quality_validate_operator), "vla_sample_manifest_validate": operator_handler(vla_sample_manifest_validate_operator), "rlbench_dataset_validate": operator_handler(rlbench_dataset_validate_operator), "maniskill_trajectory_validate": operator_handler(maniskill_trajectory_validate_operator), diff --git a/backend/operator-runner/test_main.py b/backend/operator-runner/test_main.py index 535d2b2b..87755806 100644 --- a/backend/operator-runner/test_main.py +++ b/backend/operator-runner/test_main.py @@ -279,6 +279,7 @@ class RuntimeFileContractTests(unittest.TestCase): self.assertEqual(MODULE.select_handler("LIBERO HDF5任务校验"), "libero_hdf5_task_validate") self.assertEqual(MODULE.select_handler("RoboCasa HDF5示教校验"), "robocasa_hdf5_demonstration_validate") self.assertEqual(MODULE.select_handler("RH20T 场景结构校验"), "rh20t_scene_validate") + self.assertEqual(MODULE.select_handler("RoboMIND HDF5质量校验"), "robomind_hdf5_quality_validate") self.assertEqual(MODULE.select_handler("LeRobot v3 Episode索引校验"), "lerobot_v3_episode_index_validate") self.assertEqual(MODULE.select_handler("数据集切分泄漏校验"), "dataset_split_leakage_validate") self.assertEqual(MODULE.select_handler("媒体引用可读性校验"), "media_reference_validate") @@ -3154,6 +3155,87 @@ class EmbodiedDatasetOperatorTests(unittest.TestCase): self.assertTrue(scene_payload["highFreq"]["present"]) self.assertIn("intrinsics.npy", scene_payload["calibration"]["files"]) + def test_robomind_hdf5_quality_validate_reads_real_hdf5_joint_position_and_end_effector(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: + root = pathlib.Path(temp_dir) / "robomind" / "benchmark1_0" + data_dir = root / "h5_franka_1rgb" / "bread_in_basket" / "success_episodes" / "train" / "1014_144602" / "data" + data_dir.mkdir(parents=True) + path = data_dir / "trajectory.hdf5" + with h5py.File(path, "w") as handle: + puppet = handle.create_group("puppet") + puppet.create_dataset( + "joint_position", + data=np.array( + [ + [0.0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6], + [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7], + [0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8], + ], + dtype=np.float32, + ), + ) + puppet.create_dataset( + "end_effector", + data=np.array( + [ + [0.0, 0.0, 0.0, 0.0, 0.0, 0.0], + [0.1, 0.0, 0.0, 0.0, 0.0, 0.1], + [0.2, 0.0, 0.0, 0.0, 0.0, 0.2], + ], + dtype=np.float32, + ), + ) + master = handle.create_group("master") + master.create_dataset( + "joint_position", + data=np.array( + [ + [1.0, 1.1, 1.2, 1.3, 1.4, 1.5, 1.6], + [1.1, 1.2, 1.3, 1.4, 1.5, 1.6, 1.7], + [1.2, 1.3, 1.4, 1.5, 1.6, 1.7, 1.8], + ], + dtype=np.float32, + ), + ) + rgb_images = handle.create_group("rgb_images") + rgb_images.create_dataset("front", data=np.zeros((3, 4, 4, 3), dtype=np.uint8)) + handle.create_dataset("action", data=np.ones((3, 7), dtype=np.float32)) + node_data = { + "label": "RoboMIND HDF5质量校验", + "hyperparams": [{"key": "SOURCE_PATH", "value": str(root)}], + } + + artifacts = MODULE.BUILTIN_HANDLERS["robomind_hdf5_quality_validate"]( + "run-1", "robomind", root / "out", node_data, [] + ) + + payload = MODULE.load_artifact_payload(artifacts[0]) + self.assertEqual(payload["datasetType"], "robomind-hdf5-quality-validation") + self.assertTrue(payload["valid"]) + self.assertEqual(payload["fileCount"], 1) + self.assertEqual(payload["trajectoryCount"], 1) + self.assertEqual(payload["validFileCount"], 1) + self.assertEqual(payload["jointPositionFileCount"], 1) + self.assertEqual(payload["endEffectorFileCount"], 1) + self.assertEqual(payload["stuckJointPositionCount"], 0) + self.assertEqual(payload["stuckEndEffectorCount"], 0) + self.assertEqual(payload["imageDatasetCount"], 1) + file_payload = payload["files"][0] + self.assertEqual(file_payload["benchmark"], "benchmark1_0") + self.assertEqual(file_payload["robotType"], "h5_franka_1rgb") + self.assertEqual(file_payload["taskName"], "bread_in_basket") + self.assertEqual(file_payload["split"], "train") + self.assertEqual(file_payload["episodeId"], "1014_144602") + self.assertEqual(file_payload["frameCount"], 3) + self.assertEqual(file_payload["puppetJointPosition"]["key"], "puppet/joint_position") + self.assertEqual(file_payload["puppetJointPosition"]["shape"], [3, 7]) + self.assertEqual(file_payload["puppetEndEffector"]["key"], "puppet/end_effector") + def test_hdf5_demonstration_summary_reads_real_hdf5_file(self): try: import h5py # type: ignore diff --git a/backend/operators/builtin/robomind_hdf5_quality_validate/__init__.py b/backend/operators/builtin/robomind_hdf5_quality_validate/__init__.py new file mode 100644 index 00000000..8b137891 --- /dev/null +++ b/backend/operators/builtin/robomind_hdf5_quality_validate/__init__.py @@ -0,0 +1 @@ + diff --git a/backend/operators/builtin/robomind_hdf5_quality_validate/operator.json b/backend/operators/builtin/robomind_hdf5_quality_validate/operator.json new file mode 100644 index 00000000..26c45f86 --- /dev/null +++ b/backend/operators/builtin/robomind_hdf5_quality_validate/operator.json @@ -0,0 +1,16 @@ +{ + "id": "operator-builtin-robomind-hdf5-quality-validate", + "name": "RoboMIND HDF5质量校验", + "category": "ROBOTICS", + "type": "builtin", + "runtimeHandler": "builtin_robomind_hdf5_quality_validate", + "runtimeSource": "backend/operators/builtin/robomind_hdf5_quality_validate/operator.py", + "migrationStatus": "migrated", + "acceptance": { + "requiresRealInputs": true, + "mockOnlyAccepted": false, + "simulationOnlyAccepted": false, + "hardcodedOutputAccepted": false, + "shallowOnlyAccepted": false + } +} diff --git a/backend/operators/builtin/robomind_hdf5_quality_validate/operator.py b/backend/operators/builtin/robomind_hdf5_quality_validate/operator.py new file mode 100644 index 00000000..f9cb519c --- /dev/null +++ b/backend/operators/builtin/robomind_hdf5_quality_validate/operator.py @@ -0,0 +1,412 @@ +from __future__ import annotations + +import math +import pathlib +from typing import Any, Callable + + +HelperMap = dict[str, Callable[..., Any]] +HDF5_SUFFIXES = {".h5", ".hdf5"} +PUPPET_EE_CANDIDATES = ( + "puppet/end_effector", + "puppet/end_effector_left", + "franka/end_effector", +) +PUPPET_JOINT_CANDIDATES = ( + "puppet/joint_position", + "puppet/joint_position_left", + "franka/joint_position", +) +MASTER_JOINT_CANDIDATES = ( + "master/joint_position", + "master/joint_position_left", +) +IMAGE_KEY_TOKENS = ("rgb", "image", "camera") +ACTION_KEY_TOKENS = ("action", "gripper", "command") + + +def _require_h5py(): + try: + import h5py # type: ignore + except Exception as exc: # pragma: no cover - runtime dependency check + raise RuntimeError("RoboMIND HDF5质量校验需要 h5py 读取真实 trajectory.hdf5 文件") from exc + return h5py + + +def _require_numpy(): + try: + import numpy as np # type: ignore + except Exception as exc: # pragma: no cover - runtime dependency check + raise RuntimeError("RoboMIND HDF5质量校验需要 numpy 计算关节和末端执行器质量指标") from exc + return np + + +def _positive_int(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 _positive_float(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(helpers: HelperMap, params: dict[str, str], key: str, default: bool) -> bool: + return bool(helpers["parse_bool_param"](params, key, default)) + + +def _relative(path: pathlib.Path, root: pathlib.Path) -> str: + try: + return path.relative_to(root).as_posix() + except ValueError: + return str(path) + + +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_SUFFIXES else [] + files = [item for item in sorted(source_path.rglob("*")) if item.is_file() and item.suffix.lower() in HDF5_SUFFIXES] + files.sort(key=lambda path: (0 if path.name == "trajectory.hdf5" else 1, path.as_posix())) + return files[:max_files] + + +def _parse_robomind_path(path: pathlib.Path, root: pathlib.Path) -> dict[str, str]: + parts = pathlib.PurePosixPath(_relative(path, root)).parts + robot_index = next((idx for idx, part in enumerate(parts) if part.startswith("h5_")), -1) + benchmark = parts[robot_index - 1] if robot_index > 0 and parts[robot_index - 1].startswith("benchmark") else "" + if not benchmark and robot_index == 0 and root.name.startswith("benchmark"): + benchmark = root.name + robot_type = parts[robot_index] if robot_index >= 0 else "" + task_name = parts[robot_index + 1] if robot_index >= 0 and len(parts) > robot_index + 1 else "" + split = "" + episode_id = "" + if "success_episodes" in parts: + success_index = parts.index("success_episodes") + split = parts[success_index + 1] if len(parts) > success_index + 1 else "" + episode_id = parts[success_index + 2] if len(parts) > success_index + 2 else "" + elif path.parent.name == "data": + episode_id = path.parent.parent.name + return { + "benchmark": benchmark, + "robotType": robot_type, + "taskName": task_name, + "split": split, + "episodeId": episode_id, + } + + +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_at(handle: Any, candidates: tuple[str, ...]) -> tuple[str, Any | None]: + for candidate in candidates: + if candidate in handle and hasattr(handle[candidate], "shape") and hasattr(handle[candidate], "dtype"): + return candidate, handle[candidate] + return "", None + + +def _dataset_shape(dataset: Any | None) -> list[int]: + if dataset is None or not hasattr(dataset, "shape"): + return [] + return [int(item) for item in dataset.shape] + + +def _frame_count(dataset: Any | None) -> int: + shape = _dataset_shape(dataset) + return int(shape[0]) if shape else 0 + + +def _sample_dataset(np: Any, dataset: Any, max_rows: int) -> Any: + shape = _dataset_shape(dataset) + if not shape: + return dataset[()] + frame_count = int(shape[0]) + if frame_count <= max_rows: + return dataset[()] + indices = np.unique(np.linspace(0, frame_count - 1, num=max_rows, dtype=np.int64)) + return dataset[indices, ...] + + +def _array_quality(np: Any, key: str, dataset: Any | None, tolerance: float, max_rows: int) -> dict[str, Any]: + if dataset is None: + return { + "key": "", + "exists": False, + "shape": [], + "dtype": "", + "frameCount": 0, + "sampleFrameCount": 0, + "finiteNumeric": True, + "maxRange": 0.0, + "rangePerDim": [], + "isStuck": False, + "uniqueCount": 0, + "uniqueRatio": 0.0, + } + shape = _dataset_shape(dataset) + frame_count = int(shape[0]) if shape else 1 + dtype = str(getattr(dataset, "dtype", "")) + if not np.issubdtype(dataset.dtype, np.number): + return { + "key": key, + "exists": True, + "shape": shape, + "dtype": dtype, + "frameCount": frame_count, + "sampleFrameCount": 0, + "finiteNumeric": True, + "maxRange": 0.0, + "rangePerDim": [], + "isStuck": False, + "uniqueCount": 0, + "uniqueRatio": 0.0, + } + + values = np.asarray(_sample_dataset(np, dataset, max_rows)) + if values.ndim == 0: + rows = values.reshape(1, 1) + elif values.ndim == 1: + rows = values.reshape(values.shape[0], 1) + else: + rows = values.reshape(values.shape[0], -1) + finite_numeric = bool(np.isfinite(rows).all()) + if rows.size and finite_numeric: + range_per_dim = (np.max(rows, axis=0) - np.min(rows, axis=0)).astype(float) + max_range = float(np.max(range_per_dim)) if range_per_dim.size else 0.0 + range_values = [float(item) for item in range_per_dim[:64].tolist()] + else: + max_range = 0.0 + range_values = [] + unique_count = 0 + unique_ratio = 0.0 + if rows.size: + try: + unique_count = int(np.unique(rows, axis=0).shape[0]) + except Exception: + unique_count = int(len({tuple(row.tolist()) for row in rows})) + unique_ratio = float(unique_count / max(1, int(rows.shape[0]))) + return { + "key": key, + "exists": True, + "shape": shape, + "dtype": dtype, + "frameCount": frame_count, + "sampleFrameCount": int(rows.shape[0]) if rows.ndim else 1, + "finiteNumeric": finite_numeric, + "maxRange": max_range, + "rangePerDim": range_values, + "isStuck": bool(frame_count > 1 and finite_numeric and max_range < tolerance), + "uniqueCount": unique_count, + "uniqueRatio": unique_ratio, + } + + +def _dataset_name_matches(path: str, tokens: tuple[str, ...]) -> bool: + lowered = path.lower() + return any(token in lowered for token in tokens) + + +def _summarize_file( + path: pathlib.Path, + root: pathlib.Path, + *, + max_rows: int, + ee_tolerance: float, + joint_tolerance: float, + require_joint_position: bool, + require_end_effector: bool, + require_images: bool, +) -> dict[str, Any]: + h5py = _require_h5py() + np = _require_numpy() + base_payload: dict[str, Any] = { + "path": _relative(path, root), + "fileName": path.name, + "fileSizeBytes": path.stat().st_size, + **_parse_robomind_path(path, root), + } + issues: list[dict[str, Any]] = [] + try: + with h5py.File(path, "r") as handle: + datasets = _iter_dataset_paths(handle) + if not datasets: + issues.append({"code": "missing_hdf5_datasets"}) + image_datasets = [dataset_path for dataset_path, _ in datasets if _dataset_name_matches(dataset_path, IMAGE_KEY_TOKENS)] + action_datasets = [dataset_path for dataset_path, _ in datasets if _dataset_name_matches(dataset_path, ACTION_KEY_TOKENS)] + puppet_ee_key, puppet_ee_dataset = _dataset_at(handle, PUPPET_EE_CANDIDATES) + puppet_joint_key, puppet_joint_dataset = _dataset_at(handle, PUPPET_JOINT_CANDIDATES) + master_joint_key, master_joint_dataset = _dataset_at(handle, MASTER_JOINT_CANDIDATES) + puppet_ee = _array_quality(np, puppet_ee_key, puppet_ee_dataset, ee_tolerance, max_rows) + puppet_joint = _array_quality(np, puppet_joint_key, puppet_joint_dataset, joint_tolerance, max_rows) + master_joint = _array_quality(np, master_joint_key, master_joint_dataset, joint_tolerance, max_rows) + frame_candidates = [ + _frame_count(puppet_ee_dataset), + _frame_count(puppet_joint_dataset), + _frame_count(master_joint_dataset), + ] + for dataset_path, dataset in datasets: + if _dataset_name_matches(dataset_path, IMAGE_KEY_TOKENS + ACTION_KEY_TOKENS): + frame_candidates.append(_frame_count(dataset)) + frame_count = max(frame_candidates or [0]) + if require_joint_position and not puppet_joint["exists"]: + issues.append({"code": "missing_puppet_joint_position", "candidates": list(PUPPET_JOINT_CANDIDATES)}) + if require_end_effector and not puppet_ee["exists"]: + issues.append({"code": "missing_puppet_end_effector", "candidates": list(PUPPET_EE_CANDIDATES)}) + if require_images and not image_datasets: + issues.append({"code": "missing_image_datasets"}) + for label, quality in ( + ("puppet_end_effector", puppet_ee), + ("puppet_joint_position", puppet_joint), + ("master_joint_position", master_joint), + ): + if not quality["exists"]: + continue + if not quality["finiteNumeric"]: + issues.append({"code": "non_finite_numeric_values", "dataset": quality["key"], "field": label}) + if quality["isStuck"]: + issues.append( + { + "code": "stuck_numeric_values", + "dataset": quality["key"], + "field": label, + "maxRange": quality["maxRange"], + } + ) + if frame_count <= 1: + issues.append({"code": "insufficient_frame_count", "frameCount": frame_count}) + return { + **base_payload, + "valid": not issues, + "issueCount": len(issues), + "issues": issues, + "frameCount": frame_count, + "datasetCount": len(datasets), + "imageDatasetCount": len(image_datasets), + "imageDatasets": image_datasets[:50], + "actionDatasetCount": len(action_datasets), + "actionDatasets": action_datasets[:50], + "puppetEndEffector": puppet_ee, + "puppetJointPosition": puppet_joint, + "masterJointPosition": master_joint, + } + except Exception as exc: + issues.append({"code": "unreadable_hdf5_file", "error": str(exc)}) + return { + **base_payload, + "valid": False, + "issueCount": len(issues), + "issues": issues, + "frameCount": 0, + "datasetCount": 0, + "imageDatasetCount": 0, + "imageDatasets": [], + "actionDatasetCount": 0, + "actionDatasets": [], + "puppetEndEffector": _array_quality(_require_numpy(), "", None, ee_tolerance, max_rows), + "puppetJointPosition": _array_quality(_require_numpy(), "", None, joint_tolerance, max_rows), + "masterJointPosition": _array_quality(_require_numpy(), "", None, joint_tolerance, max_rows), + } + + +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("RoboMIND HDF5质量校验需要 SOURCE_PATH 指向真实 RoboMIND trajectory.hdf5 文件或目录") + if not source_path.exists(): + raise RuntimeError(f"{source_path} does not exist") + max_files = _positive_int(params, "MAX_FILES", 100) + max_rows = _positive_int(params, "MAX_ROWS", 200000) + ee_tolerance = _positive_float(params, "EE_TOLERANCE", 0.001) + joint_tolerance = _positive_float(params, "JOINT_TOLERANCE", 0.001) + require_joint_position = _bool_param(helpers, params, "REQUIRE_JOINT_POSITION", True) + require_end_effector = _bool_param(helpers, params, "REQUIRE_END_EFFECTOR", False) + require_images = _bool_param(helpers, params, "REQUIRE_IMAGES", False) + files = _find_hdf5_files(source_path, max_files) + if not files: + raise RuntimeError(f"{source_path} does not contain RoboMIND .hdf5/.h5 files") + root = source_path.parent if source_path.is_file() else source_path + file_payloads = [ + _summarize_file( + path, + root, + max_rows=max_rows, + ee_tolerance=ee_tolerance, + joint_tolerance=joint_tolerance, + require_joint_position=require_joint_position, + require_end_effector=require_end_effector, + require_images=require_images, + ) + for path in files + ] + invalid = [item for item in file_payloads if not item["valid"]] + issues: list[dict[str, Any]] = [] + if invalid: + issues.append({"code": "invalid_robomind_trajectories", "count": len(invalid), "examples": invalid[:20]}) + payload = { + "datasetType": "robomind-hdf5-quality-validation", + "sourcePath": str(source_path), + "valid": not issues, + "issueCount": len(issues), + "issues": issues, + "fileCount": len(file_payloads), + "trajectoryCount": len(file_payloads), + "validFileCount": len(file_payloads) - len(invalid), + "invalidFileCount": len(invalid), + "frameCount": sum(int(item["frameCount"]) for item in file_payloads), + "jointPositionFileCount": sum( + 1 + for item in file_payloads + if item["puppetJointPosition"]["exists"] or item["masterJointPosition"]["exists"] + ), + "endEffectorFileCount": sum(1 for item in file_payloads if item["puppetEndEffector"]["exists"]), + "stuckJointPositionCount": sum( + 1 + for item in file_payloads + if item["puppetJointPosition"]["isStuck"] or item["masterJointPosition"]["isStuck"] + ), + "stuckEndEffectorCount": sum(1 for item in file_payloads if item["puppetEndEffector"]["isStuck"]), + "imageDatasetCount": sum(int(item["imageDatasetCount"]) for item in file_payloads), + "actionDatasetCount": sum(int(item["actionDatasetCount"]) for item in file_payloads), + "files": file_payloads, + } + return [ + helpers["write_json_artifact"]( + run_id, + node_id, + output_dir, + "robomind-hdf5-quality-validation.json", + payload, + output=helpers["output_binding"](node_data), + ) + ] diff --git a/backend/operators/builtin/robomind_hdf5_quality_validate/repository.json b/backend/operators/builtin/robomind_hdf5_quality_validate/repository.json new file mode 100644 index 00000000..d9f91bdd --- /dev/null +++ b/backend/operators/builtin/robomind_hdf5_quality_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": "RoboMIND HDF5质量校验", + "operatorId": "operator-builtin-robomind-hdf5-quality-validate", + "operatorSlug": "robomind_hdf5_quality_validate", + "preferredSlug": "robomind_hdf5_quality_validate", + "pythonImport": "backend.operators.builtin.robomind_hdf5_quality_validate.operator:run", + "repositoryId": "builtin/robomind_hdf5_quality_validate", + "runtimeDependencies": [ + "robomind-trajectory-hdf5-files", + "h5py", + "numpy" + ], + "runtimeHandler": "builtin_robomind_hdf5_quality_validate", + "runtimeSource": "backend/operators/builtin/robomind_hdf5_quality_validate/operator.py", + "schemaVersion": "droboticflow.operator.repository/v1", + "sourceRoot": "backend/operators/builtin/robomind_hdf5_quality_validate", + "strictPocAcceptanceExternals": [], + "type": "builtin" +} diff --git a/docs/operators/README.md b/docs/operators/README.md index 3fcc3746..352b3214 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:`77` +- 当前收录数量:`78` - 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) @@ -92,6 +92,7 @@ - [ROSBag 数据概览](ROSBag-数据概览.md) - [RT-X动作空间Schema校验](RT-X动作空间Schema校验.md) - [RoboCasa HDF5示教校验](RoboCasa-HDF5示教校验.md) +- [RoboMIND HDF5质量校验](RoboMIND-HDF5质量校验.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) diff --git a/docs/operators/RoboMIND-HDF5质量校验.md b/docs/operators/RoboMIND-HDF5质量校验.md new file mode 100644 index 00000000..766217bb --- /dev/null +++ b/docs/operators/RoboMIND-HDF5质量校验.md @@ -0,0 +1,121 @@ +# RoboMIND HDF5质量校验 + +- 算子 ID:`operator-builtin-robomind-hdf5-quality-validate` +- 类别:`ROBOTICS` +- 类型:`builtin` + +## 作用 + +读取真实 RoboMIND trajectory.hdf5,校验 puppet/master joint_position、puppet/end_effector 卡死、非有限值、帧数和多模态字段。 + +## 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 RoboMIND、Multi-embodiment Intelligence、真实遥操作 VLA 训练数据,读取 h5_franka_1rgb/h5_franka_3rgb/h5_ur_1rgb/h5_agilex_3rgb/h5_tienkung 等目录下的 trajectory.hdf5,解析 robot/task/split/episode 元信息,检查 puppet/end_effector、puppet/joint_position、master/joint_position、franka/* 候选字段的 shape、finite 数值、unique ratio 和卡死状态,并统计 RGB/image/camera 与 action/gripper/command 数据集。 + +## 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/robomind_hdf5_quality_validate/operator.py` 中的 `run(...)`,平台 runtimeHandler 为 `builtin_robomind_hdf5_quality_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[RoboMIND HDF5质量校验] + B --> C[artifact / stage manifest / 外部副作用] +``` + +## 输入端口详解 + +- `in-1` / **input** / `json` / `bus`:RoboMIND trajectory.hdf5 文件或包含 h5_* 轨迹目录的根目录 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`:RoboMIND HDF5 轨迹质量校验结果 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。RoboMIND trajectory.hdf5 文件、data 目录或包含 benchmark/h5_* 的根目录 通常建议在工作流中显式配置。 +- `MAX_FILES`:类型 `string`,默认值 `100`。最多校验多少个 .hdf5/.h5 轨迹文件,优先读取 trajectory.hdf5 可直接使用默认值,也可在节点级覆盖。 +- `MAX_ROWS`:类型 `string`,默认值 `200000`。每个数值数组最多采样多少帧用于 range 和 unique ratio 计算 可直接使用默认值,也可在节点级覆盖。 +- `EE_TOLERANCE`:类型 `string`,默认值 `0.001`。末端执行器数据 max(range_per_dim) 小于该阈值时判定卡死 可直接使用默认值,也可在节点级覆盖。 +- `JOINT_TOLERANCE`:类型 `string`,默认值 `0.001`。关节数据 max(range_per_dim) 小于该阈值时判定卡死 可直接使用默认值,也可在节点级覆盖。 +- `REQUIRE_JOINT_POSITION`:类型 `boolean`,默认值 `true`。是否要求存在 puppet/joint_position、puppet/joint_position_left 或 franka/joint_position 可直接使用默认值,也可在节点级覆盖。 +- `REQUIRE_END_EFFECTOR`:类型 `boolean`,默认值 `false`。是否要求存在 puppet/end_effector、puppet/end_effector_left 或 franka/end_effector;默认仅在存在时检查质量 可直接使用默认值,也可在节点级覆盖。 +- `REQUIRE_IMAGES`:类型 `boolean`,默认值 `false`。是否要求 HDF5 中存在 rgb/image/camera 相关观测数据集 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": "RoboMIND HDF5质量校验", + "operatorId": "operator-builtin-robomind-hdf5-quality-validate", + "type": "builtin", + "category": "ROBOTICS", + "hyperparams": { + "SOURCE_PATH": "<按节点填写>", + "MAX_FILES": "100", + "MAX_ROWS": "200000", + "EE_TOLERANCE": "0.001", + "JOINT_TOLERANCE": "0.001", + "REQUIRE_JOINT_POSITION": "true", + "REQUIRE_END_EFFECTOR": "false", + "REQUIRE_IMAGES": "false" + } +} +``` + +## 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 6dcd1306..062bb48f 100644 --- a/internal/api/builtin_operators_catalog.json +++ b/internal/api/builtin_operators_catalog.json @@ -1928,6 +1928,84 @@ "runtimeSource": "backend/operators/builtin/rh20t_scene_validate/operator.py", "runtimeHandler": "builtin_rh20t_scene_validate" }, + { + "id": "operator-builtin-robomind-hdf5-quality-validate", + "name": "RoboMIND HDF5质量校验", + "type": "builtin", + "category": "ROBOTICS", + "description": "读取真实 RoboMIND trajectory.hdf5,校验 puppet/master joint_position、puppet/end_effector 卡死、非有限值、帧数和多模态字段。", + "inputs": [ + { + "id": "in-1", + "name": "input", + "type": "json", + "description": "RoboMIND trajectory.hdf5 文件或包含 h5_* 轨迹目录的根目录", + "transmission": "bus" + } + ], + "outputs": [ + { + "id": "out-1", + "name": "validation", + "type": "json", + "description": "RoboMIND HDF5 轨迹质量校验结果", + "transmission": "bus" + } + ], + "skillDefinition": "面向 RoboMIND、Multi-embodiment Intelligence、真实遥操作 VLA 训练数据,读取 h5_franka_1rgb/h5_franka_3rgb/h5_ur_1rgb/h5_agilex_3rgb/h5_tienkung 等目录下的 trajectory.hdf5,解析 robot/task/split/episode 元信息,检查 puppet/end_effector、puppet/joint_position、master/joint_position、franka/* 候选字段的 shape、finite 数值、unique ratio 和卡死状态,并统计 RGB/image/camera 与 action/gripper/command 数据集。", + "hyperparams": [ + { + "key": "SOURCE_PATH", + "value": "", + "description": "RoboMIND trajectory.hdf5 文件、data 目录或包含 benchmark/h5_* 的根目录", + "type": "string" + }, + { + "key": "MAX_FILES", + "value": "100", + "description": "最多校验多少个 .hdf5/.h5 轨迹文件,优先读取 trajectory.hdf5", + "type": "string" + }, + { + "key": "MAX_ROWS", + "value": "200000", + "description": "每个数值数组最多采样多少帧用于 range 和 unique ratio 计算", + "type": "string" + }, + { + "key": "EE_TOLERANCE", + "value": "0.001", + "description": "末端执行器数据 max(range_per_dim) 小于该阈值时判定卡死", + "type": "string" + }, + { + "key": "JOINT_TOLERANCE", + "value": "0.001", + "description": "关节数据 max(range_per_dim) 小于该阈值时判定卡死", + "type": "string" + }, + { + "key": "REQUIRE_JOINT_POSITION", + "value": "true", + "description": "是否要求存在 puppet/joint_position、puppet/joint_position_left 或 franka/joint_position", + "type": "boolean" + }, + { + "key": "REQUIRE_END_EFFECTOR", + "value": "false", + "description": "是否要求存在 puppet/end_effector、puppet/end_effector_left 或 franka/end_effector;默认仅在存在时检查质量", + "type": "boolean" + }, + { + "key": "REQUIRE_IMAGES", + "value": "false", + "description": "是否要求 HDF5 中存在 rgb/image/camera 相关观测数据集", + "type": "boolean" + } + ], + "runtimeSource": "backend/operators/builtin/robomind_hdf5_quality_validate/operator.py", + "runtimeHandler": "builtin_robomind_hdf5_quality_validate" + }, { "id": "operator-builtin-hdf5-demonstration-summary", "name": "HDF5 示教数据概览", diff --git a/scripts/generate_operator_repository_manifests.py b/scripts/generate_operator_repository_manifests.py index 41e2670d..f4d32913 100644 --- a/scripts/generate_operator_repository_manifests.py +++ b/scripts/generate_operator_repository_manifests.py @@ -65,6 +65,7 @@ RUNTIME_DEPENDENCIES: dict[str, list[str]] = { "rgbd_frame_pair_validate": ["rgb-images", "depth-images", "opencv-python-headless", "numpy", "pyyaml"], "rlds_episode_summary": ["rlds-json-or-tfds-files"], "rh20t_scene_validate": ["rh20t-scene-directory", "numpy", "npy-files", "image-video-files"], + "robomind_hdf5_quality_validate": ["robomind-trajectory-hdf5-files", "h5py", "numpy"], "robomimic_hdf5_validate": ["robomimic-hdf5-files", "h5py", "numpy"], "robocasa_hdf5_demonstration_validate": ["robocasa-hdf5-demo-files", "h5py", "numpy"], "robot_model_asset_validate": ["urdf-files", "mjcf-files", "mesh-assets"],