File size: 17,190 Bytes
42d9709 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 | # SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import os
import shutil
from dataclasses import dataclass, field
from enum import Enum
from pathlib import Path
from typing import Optional
class EvalTaskConfig(Enum):
NUTPOURING = (
"Isaac-NutPour-GR1T2-ClosedLoop-v0",
"/home/gr00t/GR00T-N1-2B-tuned-Nut-Pouring-task",
(
"Pick up the beaker and tilt it to pour out 1 metallic nut into the bowl. Pick up the bowl and place it on"
" the metallic measuring scale."
),
"nut_pouring_task.hdf5",
0 # 1 is reserved for data validity check, following GR00T-N1 guidelines.
)
PIPESORTING = (
"Isaac-ExhaustPipe-GR1T2-ClosedLoop-v0",
"/home/gr00t/GR00T-N1-2B-tuned-Exhaust-Pipe-Sorting-task",
"Pick up the blue pipe and place it into the blue bin.",
"exhaust_pipe_sorting_task.hdf5",
2 # 1 is reserved for data validity check, following GR00T-N1 guidelines.
)
PICKPLACE_LARGE = (
"Isaac-PickPlace-Camera-G1-v0",
"~/IsaacLabEvalTasks/datasets/Isaac-PickPlace-Camera-G1-v0",
"Pick up the steering wheel and place it into the basket.",
"generated_dataset_pick_place_camera_g1.hdf5",
3 # 1 is reserved for data validity check, following GR00T-N1 guidelines.
)
APPLE_LARGE = (
"Isaac-Apple-PickPlace-G1-v0",
"~/IsaacLabEvalTasks/datasets/Isaac-Apple-PickPlace-G1-v0",
"Pick up the apple and place it on the plate.",
"apple_pick_place_generated.hdf5",
4
)
# Pick and place apple 5 teleop demonstrations
APPLE_5 = (
"Isaac-Apple-PickPlace-G1-v0",
"~/isaaclabevaltasks/datasets",
"Pick up the apple and place it on the plate.",
"apple_pick_place_annotated.hdf5",
5
)
# Pick and place apple 20 generated demonstrations
APPLE_20 = ( # Specified by: --task_name apple_20
"Isaac-Apple-PickPlace-G1-v0",
"~/isaaclabevaltasks/datasets", # Specified by: --root_dir <DIR>
"Pick up the apple and place it on the plate.",
"apple_pick_place_generated_small.hdf5",
5
)
# === Define and new task here ===
STEERING_WHEEL = (
"Isaac-PickPlace-Camera-G1-Mimic-v0",
"~/isaaclab/datasets",
"Pick up the steering wheel and place it on the basket.",
"steering_wheel_generated.hdf5",
6
)
def __init__(self, task: str, model_path: str, language_instruction: str, hdf5_name: str, task_index: int):
self.task = task
self.model_path = model_path
self.language_instruction = language_instruction
self.hdf5_name = hdf5_name
assert task_index != 1, "task_index must not be 1. (Use 0 for nutpouring, 2 for exhaustpipe, etc.)"
self.task_index = task_index
@dataclass
class Gr00tN1ClosedLoopArguments:
# Simulation specific parameters
headless: bool = field(
default=False, metadata={"description": "Whether to run the simulator in headless (no GUI) mode."}
)
num_envs: int = field(default=10, metadata={"description": "Number of environments to run in parallel."})
enable_pinocchio: bool = field(
default=True,
metadata={
"description": (
"Whether to use Pinocchio for physics simulation. Required for NutPouring and ExhaustPipe tasks."
)
},
)
record_camera: bool = field(
default=False,
metadata={"description": "Whether to record the camera images as videos during evaluation."},
)
record_video_output_path: str = field(
default="videos/",
metadata={"description": "Path to save the recorded videos."},
)
# model specific parameters
task_name: str = field(
default="nutpouring", metadata={"description": "Short name of the task to run (e.g., nutpouring, exhaustpipe)."}
)
task: str = field(default="", metadata={"description": "Full task name for the gym-registered environment."})
language_instruction: str = field(
default="", metadata={"description": "Instruction given to the policy in natural language."}
)
model_path: str = field(default="", metadata={"description": "Full path to the tuned model checkpoint directory."})
action_horizon: int = field(
default=16, metadata={"description": "Number of actions in the policy's predictionhorizon."}
)
embodiment_tag: str = field(
default="g1",
metadata={
"description": (
"Identifier for the robot embodiment used in the policy inference (e.g., 'g1' or 'new_embodiment')."
)
},
)
denoising_steps: int = field(
default=4, metadata={"description": "Number of denoising steps used in the policy inference."}
)
data_config: str = field(
default="g1", metadata={"description": "Name of the data configuration to use for the policy."}
)
original_image_size: tuple[int, int, int] = field(
default=(160, 256, 3), metadata={"description": "Original size of input images as (height, width, channels)."}
)
target_image_size: tuple[int, int, int] = field(
default=(256, 256, 3),
metadata={"description": "Target size for images after resizing and padding as (height, width, channels)."},
)
gr00t_joints_config_path: Path = field(
default=Path(__file__).parent.resolve() / "gr00t_g1" / "gr00t_joint_space.yaml",
metadata={"description": "Path to the YAML file specifying the joint ordering configuration for GR00T policy."},
)
# robot (G1) simulation specific parameters
action_joints_config_path: Path = field(
default=Path(__file__).parent.resolve() / "g1" / "action_joint_space.yaml",
metadata={
"description": (
"Path to the YAML file specifying the joint ordering configuration for G1 action space in Lab."
)
},
)
state_joints_config_path: Path = field(
default=Path(__file__).parent.resolve() / "g1" / "state_joint_space.yaml",
metadata={
"description": (
"Path to the YAML file specifying the joint ordering configuration for G1 state space in Lab."
)
},
)
# Default to GPU policy and CPU physics simulation
policy_device: str = field(
default="cuda", metadata={"description": "Device to run the policy model on (e.g., 'cuda' or 'cpu')."}
)
simulation_device: str = field(
default="cpu", metadata={"description": "Device to run the physics simulation on (e.g., 'cpu' or 'cuda')."}
)
# Evaluation parameters
max_num_rollouts: int = field(
default=100, metadata={"description": "Maximum number of rollouts to perform during evaluation."}
)
checkpoint_name: str = field(
default="gr00t-n1-2b-tuned", metadata={"description": "Name of the model checkpoint used for evaluation."}
)
eval_file_path: Optional[str] = field(
default=None, metadata={"description": "Path to the file where evaluation results will be saved."}
)
# Closed loop specific parameters
num_feedback_actions: int = field(
default=16,
metadata={
"description": "Number of feedback actions to execute per rollout (can be less than action_horizon)."
},
)
rollout_length: int = field(default=30, metadata={"description": "Number of steps in each rollout episode."})
seed: int = field(default=10, metadata={"description": "Random seed for reproducibility."})
def __post_init__(self):
# Populate fields from enum based on task_name
if self.task_name.upper() not in EvalTaskConfig.__members__:
raise ValueError(f"task_name must be one of: {', '.join(EvalTaskConfig.__members__.keys())}")
config = EvalTaskConfig[self.task_name.upper()]
if self.task == "":
self.task = config.task
if self.model_path == "":
self.model_path = config.model_path
if self.language_instruction == "":
self.language_instruction = config.language_instruction
# If model path is relative, return error
if not os.path.isabs(self.model_path):
raise ValueError("model_path must be an absolute path. Do not use relative paths.")
assert (
self.num_feedback_actions <= self.action_horizon
), "num_feedback_actions must be less than or equal to action_horizon"
# assert all paths exist
assert Path(self.gr00t_joints_config_path).exists(), "gr00t_joints_config_path does not exist"
assert Path(self.action_joints_config_path).exists(), "action_joints_config_path does not exist"
assert Path(self.state_joints_config_path).exists(), "state_joints_config_path does not exist"
assert Path(self.model_path).exists(), "model_path does not exist."
# embodiment_tag
assert self.embodiment_tag in [
"g1",
"new_embodiment",
], "embodiment_tag must be one of the following: " + ", ".join(["g1", "new_embodiment"])
@dataclass
class Gr00tN1DatasetConfig:
# Datasets & task specific parameters
data_root: Path = field(
default=Path("/mnt/datab/PhysicalAI-GR00T-Tuned-Tasks"),
metadata={"description": "Root directory for all data storage."},
)
task_name: str = field(
default="nutpouring", metadata={"description": "Short name of the task to run (e.g., nutpouring, exhaustpipe)."}
)
language_instruction: str = field(
default="", metadata={"description": "Instruction given to the policy in natural language."}
)
hdf5_name: str = field(default="", metadata={"description": "Name of the HDF5 file to use for the dataset."})
# Mimic-generated HDF5 datafield
state_name_sim: str = field(
default="robot_joint_pos", metadata={"description": "Name of the state in the HDF5 file."}
)
action_name_sim: str = field(
default="processed_actions", metadata={"description": "Name of the action in the HDF5 file."}
)
pov_cam_name_sim: str = field(
default="robot_pov_cam", metadata={"description": "Name of the POV camera in the HDF5 file."}
)
# Gr00t-LeRobot datafield
state_name_lerobot: str = field(
default="observation.state", metadata={"description": "Name of the state in the LeRobot file."}
)
action_name_lerobot: str = field(
default="action", metadata={"description": "Name of the action in the LeRobot file."}
)
video_name_lerobot: str = field(
default="observation.images.ego_view", metadata={"description": "Name of the video in the LeRobot file."}
)
task_description_lerobot: str = field(
default="annotation.human.action.task_description",
metadata={"description": "Name of the task description in the LeRobot file."},
)
valid_lerobot: str = field(
default="annotation.human.action.valid", metadata={"description": "Name of the validity in the LeRobot file."}
)
# Parquet
chunks_size: int = field(default=1000, metadata={"description": "Number of episodes per data chunk."})
# mp4 video
fps: int = field(default=20, metadata={"description": "Frames per second for video recording."})
# Metadata files
data_path: str = field(
default="data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet",
metadata={"description": "Template path for storing episode data files."},
)
video_path: str = field(
default="videos/chunk-{episode_chunk:03d}/{video_key}/episode_{episode_index:06d}.mp4",
metadata={"description": "Template path for storing episode video files."},
)
modality_template_path: Path = field(
default=Path(__file__).parent.resolve() / "gr00t_g1" / "modality.json",
metadata={"description": "Path to the modality template JSON file."},
)
modality_fname: str = field(
default="modality.json", metadata={"description": "Filename for the modality JSON file."}
)
episodes_fname: str = field(
default="episodes.jsonl", metadata={"description": "Filename for the episodes JSONL file."}
)
tasks_fname: str = field(default="tasks.jsonl", metadata={"description": "Filename for the tasks JSONL file."})
info_template_path: Path = field(
default=Path(__file__).parent.resolve() / "gr00t_g1" / "info.json",
metadata={"description": "Path to the info template JSON file."},
)
info_fname: str = field(default="info.json", metadata={"description": "Filename for the info JSON file."})
# GR00T policy specific parameters
gr00t_joints_config_path: Path = field(
default=Path(__file__).parent.resolve() / "gr00t_g1" / "gr00t_joint_space.yaml",
metadata={"description": "Path to the YAML file specifying the joint ordering configuration for GR00T policy."},
)
robot_type: str = field(
default="g1", metadata={"description": "Type of robot embodiment used in the policy fine-tuning."}
)
# robot (G1) simulation specific parameters
action_joints_config_path: Path = field(
default=Path(__file__).parent.resolve() / "g1" / "action_joint_space.yaml",
metadata={
"description": (
"Path to the YAML file specifying the joint ordering configuration for G1 action space in Lab."
)
},
)
state_joints_config_path: Path = field(
default=Path(__file__).parent.resolve() / "g1" / "state_joint_space.yaml",
metadata={
"description": (
"Path to the YAML file specifying the joint ordering configuration for G1 state space in Lab."
)
},
)
original_image_size: tuple[int, int, int] = field(
default=(160, 256, 3), metadata={"description": "Original size of input images as (height, width, channels)."}
)
target_image_size: tuple[int, int, int] = field(
default=(256, 256, 3), metadata={"description": "Target size for images after resizing and padding."}
)
hdf5_file_path: Path = field(init=False)
lerobot_data_dir: Path = field(init=False)
task_index: int = field(init=False) # task index for the task description in LeRobot file
def __post_init__(self):
# Populate fields from enum based on task_name
if self.task_name.upper() not in EvalTaskConfig.__members__:
raise ValueError(f"task_name must be one of: {', '.join(EvalTaskConfig.__members__.keys())}")
config = EvalTaskConfig[self.task_name.upper()]
self.language_instruction = config.language_instruction
self.hdf5_name = config.hdf5_name
self.task_index = config.task_index
self.hdf5_file_path = self.data_root / self.hdf5_name
self.lerobot_data_dir = self.data_root / self.hdf5_name.replace(".hdf5", "") / "lerobot"
# Assert all paths exist
assert self.hdf5_file_path.exists(), "hdf5_file_path does not exist"
assert Path(self.gr00t_joints_config_path).exists(), "gr00t_joints_config_path does not exist"
assert Path(self.action_joints_config_path).exists(), "action_joints_config_path does not exist"
assert Path(self.state_joints_config_path).exists(), "state_joints_config_path does not exist"
assert Path(self.info_template_path).exists(), "info_template_path does not exist"
assert Path(self.modality_template_path).exists(), "modality_template_path does not exist"
# if lerobot_data_dir not empty, throw a warning and remove
if self.lerobot_data_dir.exists():
print(f"Warning: lerobot_data_dir {self.lerobot_data_dir} already exists. Removing it.")
# remove directory contents and the directory itself using shutil
shutil.rmtree(self.lerobot_data_dir)
# Prepare data keys for mimic-generated hdf5 file
self.hdf5_keys = {
"state": self.state_name_sim,
"action": self.action_name_sim,
}
# Prepare data keys for LeRobot file
self.lerobot_keys = {
"state": self.state_name_lerobot,
"action": self.action_name_lerobot,
"video": self.video_name_lerobot,
"annotation": (
self.task_description_lerobot,
self.valid_lerobot,
),
}
|