tangger/lerobot
1
0
Fork 1
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Add typos checks (#770)

Browse Source
This commit is contained in:
Simon Alibert
2025-02-25 23:51:15 +01:00
committed by GitHub
parent 8699a28be0
commit a1809ad3de
47 changed files with 114 additions and 82 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
lerobot/common/datasets/compute_stats.py
View File

@@ -92,7 +92,7 @@ def compute_episode_stats(episode_data: dict[str, list[str] | np.ndarray], featu
axes_to_reduce = (0, 2, 3) # keep channel dim
keepdims = True
else:
ep_ft_array = data # data is alreay a np.ndarray
ep_ft_array = data # data is already a np.ndarray
axes_to_reduce = 0 # compute stats over the first axis
keepdims = data.ndim == 1 # keep as np.array

6
lerobot/common/datasets/lerobot_dataset.py
View File

@@ -226,7 +226,7 @@ class LeRobotDatasetMetadata:
def add_task(self, task: str):
"""
Given a task in natural language, add it to the dictionnary of tasks.
Given a task in natural language, add it to the dictionary of tasks.
"""
if task in self.task_to_task_index:
raise ValueError(f"The task '{task}' already exists and can't be added twice.")
@@ -389,7 +389,7 @@ class LeRobotDataset(torch.utils.data.Dataset):
- info contains various information about the dataset like shapes, keys, fps etc.
- stats stores the dataset statistics of the different modalities for normalization
- tasks contains the prompts for each task of the dataset, which can be used for
task-conditionned training.
task-conditioned training.
- hf_dataset (from datasets.Dataset), which will read any values from parquet files.
- videos (optional) from which frames are loaded to be synchronous with data from parquet files.
@@ -848,7 +848,7 @@ class LeRobotDataset(torch.utils.data.Dataset):
episode_buffer["index"] = np.arange(self.meta.total_frames, self.meta.total_frames + episode_length)
episode_buffer["episode_index"] = np.full((episode_length,), episode_index)
# Add new tasks to the tasks dictionnary
# Add new tasks to the tasks dictionary
for task in episode_tasks:
task_index = self.meta.get_task_index(task)
if task_index is None:

2
lerobot/common/datasets/push_dataset_to_hub/_download_raw.py
View File

@@ -152,7 +152,7 @@ def download_raw(raw_dir: Path, repo_id: str):
stacklevel=1,
)
# Send warning if raw_dir isn't well formated
# Send warning if raw_dir isn't well formatted
if raw_dir.parts[-2] != user_id or raw_dir.parts[-1] != dataset_id:
warnings.warn(
f"""`raw_dir` ({raw_dir}) doesn't contain a community or user id `/` the name of the dataset that

8
lerobot/common/datasets/push_dataset_to_hub/dora_parquet_format.py
View File

@@ -68,9 +68,9 @@ def load_from_raw(raw_dir: Path, videos_dir: Path, fps: int, video: bool, episod
modality_df,
on="timestamp_utc",
# "nearest" is the best option over "backward", since the latter can desynchronizes camera timestamps by
# matching timestamps that are too far appart, in order to fit the backward constraints. It's not the case for "nearest".
# matching timestamps that are too far apart, in order to fit the backward constraints. It's not the case for "nearest".
# However, note that "nearest" might synchronize the reference camera with other cameras on slightly future timestamps.
# are too far appart.
# are too far apart.
direction="nearest",
tolerance=pd.Timedelta(f"{1 / fps} seconds"),
)
@@ -126,7 +126,7 @@ def load_from_raw(raw_dir: Path, videos_dir: Path, fps: int, video: bool, episod
videos_dir.parent.mkdir(parents=True, exist_ok=True)
videos_dir.symlink_to((raw_dir / "videos").absolute())
# sanity check the video paths are well formated
# sanity check the video paths are well formatted
for key in df:
if "observation.images." not in key:
continue
@@ -143,7 +143,7 @@ def load_from_raw(raw_dir: Path, videos_dir: Path, fps: int, video: bool, episod
# it is the case for video_frame dictionary = [{"path": ..., "timestamp": ...}]
data_dict[key] = [video_frame[0] for video_frame in df[key].values]
# sanity check the video path is well formated
# sanity check the video path is well formatted
video_path = videos_dir.parent / data_dict[key][0]["path"]
if not video_path.exists():
raise ValueError(f"Video file not found in {video_path}")

2
lerobot/common/datasets/push_dataset_to_hub/openx_rlds_format.py
View File

@@ -17,7 +17,7 @@
For all datasets in the RLDS format.
For https://github.com/google-deepmind/open_x_embodiment (OPENX) datasets.
NOTE: You need to install tensorflow and tensorflow_datsets before running this script.
NOTE: You need to install tensorflow and tensorflow_datasets before running this script.
Example:
python lerobot/scripts/push_dataset_to_hub.py \

8
lerobot/common/datasets/utils.py
View File

@@ -222,7 +222,7 @@ def load_episodes(local_dir: Path) -> dict:
def write_episode_stats(episode_index: int, episode_stats: dict, local_dir: Path):
# We wrap episode_stats in a dictionnary since `episode_stats["episode_index"]`
# We wrap episode_stats in a dictionary since `episode_stats["episode_index"]`
# is a dictionary of stats and not an integer.
episode_stats = {"episode_index": episode_index, "stats": serialize_dict(episode_stats)}
append_jsonlines(episode_stats, local_dir / EPISODES_STATS_PATH)
@@ -445,10 +445,10 @@ def get_episode_data_index(
if episodes is not None:
episode_lengths = {ep_idx: episode_lengths[ep_idx] for ep_idx in episodes}
cumulative_lenghts = list(accumulate(episode_lengths.values()))
cumulative_lengths = list(accumulate(episode_lengths.values()))
return {
"from": torch.LongTensor([0] + cumulative_lenghts[:-1]),
"to": torch.LongTensor(cumulative_lenghts),
"from": torch.LongTensor([0] + cumulative_lengths[:-1]),
"to": torch.LongTensor(cumulative_lengths),
}

2
lerobot/common/datasets/v2/batch_convert_dataset_v1_to_v2.py
View File

@@ -31,6 +31,7 @@ from lerobot.common.robot_devices.robots.configs import AlohaRobotConfig
LOCAL_DIR = Path("data/")
# spellchecker:off
ALOHA_MOBILE_INFO = {
"robot_config": AlohaRobotConfig(),
"license": "mit",
@@ -856,6 +857,7 @@ DATASETS = {
}""").lstrip(),
},
}
# spellchecker:on
def batch_convert():

2
lerobot/common/datasets/v2/convert_dataset_v1_to_v2.py
View File

@@ -17,7 +17,7 @@
"""
This script will help you convert any LeRobot dataset already pushed to the hub from codebase version 1.6 to
2.0. You will be required to provide the 'tasks', which is a short but accurate description in plain English
for each of the task performed in the dataset. This will allow to easily train models with task-conditionning.
for each of the task performed in the dataset. This will allow to easily train models with task-conditioning.
We support 3 different scenarios for these tasks (see instructions below):
1. Single task dataset: all episodes of your dataset have the same single task.

2
lerobot/common/datasets/video_utils.py
View File

@@ -73,7 +73,7 @@ def decode_video_frames_torchvision(
last_ts = max(timestamps)
# access closest key frame of the first requested frame
# Note: closest key frame timestamp is usally smaller than `first_ts` (e.g. key frame can be the first frame of the video)
# Note: closest key frame timestamp is usually smaller than `first_ts` (e.g. key frame can be the first frame of the video)
# for details on what `seek` is doing see: https://pyav.basswood-io.com/docs/stable/api/container.html?highlight=inputcontainer#av.container.InputContainer.seek
reader.seek(first_ts, keyframes_only=keyframes_only)

4
lerobot/common/envs/factory.py
View File

@@ -37,12 +37,12 @@ def make_env(cfg: EnvConfig, n_envs: int = 1, use_async_envs: bool = False) -> g
Args:
cfg (EnvConfig): the config of the environment to instantiate.
n_envs (int, optional): The number of parallelized env to return. Defaults to 1.
use_async_envs (bool, optional): Wether to return an AsyncVectorEnv or a SyncVectorEnv. Defaults to
use_async_envs (bool, optional): Whether to return an AsyncVectorEnv or a SyncVectorEnv. Defaults to
False.
Raises:
ValueError: if n_envs < 1
ModuleNotFoundError: If the requested env package is not intalled
ModuleNotFoundError: If the requested env package is not installed
Returns:
gym.vector.VectorEnv: The parallelized gym.env instance.

2
lerobot/common/policies/act/configuration_act.py
View File

@@ -64,7 +64,7 @@ class ACTConfig(PreTrainedConfig):
output_normalization_modes: Similar dictionary as `normalize_input_modes`, but to unnormalize to the
original scale. Note that this is also used for normalizing the training targets.
vision_backbone: Name of the torchvision resnet backbone to use for encoding images.
pretrained_backbone_weights: Pretrained weights from torchvision to initalize the backbone.
pretrained_backbone_weights: Pretrained weights from torchvision to initialize the backbone.
`None` means no pretrained weights.
replace_final_stride_with_dilation: Whether to replace the ResNet's final 2x2 stride with a dilated
convolution.

4
lerobot/common/policies/diffusion/configuration_diffusion.py
View File

@@ -68,7 +68,7 @@ class DiffusionConfig(PreTrainedConfig):
within the image size. If None, no cropping is done.
crop_is_random: Whether the crop should be random at training time (it's always a center crop in eval
mode).
pretrained_backbone_weights: Pretrained weights from torchvision to initalize the backbone.
pretrained_backbone_weights: Pretrained weights from torchvision to initialize the backbone.
`None` means no pretrained weights.
use_group_norm: Whether to replace batch normalization with group normalization in the backbone.
The group sizes are set to be about 16 (to be precise, feature_dim // 16).
@@ -99,7 +99,7 @@ class DiffusionConfig(PreTrainedConfig):
num_inference_steps: Number of reverse diffusion steps to use at inference time (steps are evenly
spaced). If not provided, this defaults to be the same as `num_train_timesteps`.
do_mask_loss_for_padding: Whether to mask the loss when there are copy-padded actions. See
`LeRobotDataset` and `load_previous_and_future_frames` for mor information. Note, this defaults
`LeRobotDataset` and `load_previous_and_future_frames` for more information. Note, this defaults
to False as the original Diffusion Policy implementation does the same.
"""

2
lerobot/common/policies/pi0/conversion_scripts/convert_pi0_to_hf_lerobot.py
View File

@@ -2,7 +2,7 @@
Convert pi0 parameters from Jax to Pytorch
Follow [README of openpi](https://github.com/Physical-Intelligence/openpi) to create a new environment
and install the required librairies.
and install the required libraries.
```bash
cd ~/code/openpi

4
lerobot/common/policies/tdmpc/configuration_tdmpc.py
View File

@@ -76,7 +76,7 @@ class TDMPCConfig(PreTrainedConfig):
n_pi_samples: Number of samples to draw from the policy / world model rollout every CEM iteration. Can
be zero.
uncertainty_regularizer_coeff: Coefficient for the uncertainty regularization used when estimating
trajectory values (this is the λ coeffiecient in eqn 4 of FOWM).
trajectory values (this is the λ coefficient in eqn 4 of FOWM).
n_elites: The number of elite samples to use for updating the gaussian parameters every CEM iteration.
elite_weighting_temperature: The temperature to use for softmax weighting (by trajectory value) of the
elites, when updating the gaussian parameters for CEM.
@@ -165,7 +165,7 @@ class TDMPCConfig(PreTrainedConfig):
"""Input validation (not exhaustive)."""
if self.n_gaussian_samples <= 0:
raise ValueError(
f"The number of guassian samples for CEM should be non-zero. Got `{self.n_gaussian_samples=}`"
f"The number of gaussian samples for CEM should be non-zero. Got `{self.n_gaussian_samples=}`"
)
if self.normalization_mapping["ACTION"] is not NormalizationMode.MIN_MAX:
raise ValueError(

2
lerobot/common/policies/vqbet/configuration_vqbet.py
View File

@@ -66,7 +66,7 @@ class VQBeTConfig(PreTrainedConfig):
within the image size. If None, no cropping is done.
crop_is_random: Whether the crop should be random at training time (it's always a center crop in eval
mode).
pretrained_backbone_weights: Pretrained weights from torchvision to initalize the backbone.
pretrained_backbone_weights: Pretrained weights from torchvision to initialize the backbone.
`None` means no pretrained weights.
use_group_norm: Whether to replace batch normalization with group normalization in the backbone.
The group sizes are set to be about 16 (to be precise, feature_dim // 16).

4
lerobot/common/policies/vqbet/modeling_vqbet.py
View File

@@ -485,7 +485,7 @@ class VQBeTHead(nn.Module):
def forward(self, x, **kwargs) -> dict:
# N is the batch size, and T is number of action query tokens, which are process through same GPT
N, T, _ = x.shape
# we calculate N and T side parallely. Thus, the dimensions would be
# we calculate N and T side parallelly. Thus, the dimensions would be
# (batch size * number of action query tokens, action chunk size, action dimension)
x = einops.rearrange(x, "N T WA -> (N T) WA")
@@ -772,7 +772,7 @@ class VqVae(nn.Module):
Encoder and decoder are MLPs consisting of an input, output layer, and hidden layer, respectively.
The vq_layer uses residual VQs.
This class contains functions for training the encoder and decoder along with the residual VQ layer (for trainign phase 1),
This class contains functions for training the encoder and decoder along with the residual VQ layer (for training phase 1),
as well as functions to help BeT training part in training phase 2.
"""

10
lerobot/common/policies/vqbet/vqbet_utils.py
View File

@@ -38,7 +38,7 @@ from lerobot.common.policies.vqbet.configuration_vqbet import VQBeTConfig
This file is part of a VQ-BeT that utilizes code from the following repositories:
- Vector Quantize PyTorch code is licensed under the MIT License:
Origianl source: https://github.com/lucidrains/vector-quantize-pytorch
Original source: https://github.com/lucidrains/vector-quantize-pytorch
- nanoGPT part is an adaptation of Andrej Karpathy's nanoGPT implementation in PyTorch.
Original source: https://github.com/karpathy/nanoGPT
@@ -289,7 +289,7 @@ class GPT(nn.Module):
This file is a part for Residual Vector Quantization that utilizes code from the following repository:
- Phil Wang's vector-quantize-pytorch implementation in PyTorch.
Origianl source: https://github.com/lucidrains/vector-quantize-pytorch
Original source: https://github.com/lucidrains/vector-quantize-pytorch
- The vector-quantize-pytorch code is licensed under the MIT License:
@@ -1349,9 +1349,9 @@ class EuclideanCodebook(nn.Module):
# calculate distributed variance
variance_numer = reduce((data - batch_mean) ** 2, "h n d -> h 1 d", "sum")
distributed.all_reduce(variance_numer)
batch_variance = variance_numer / num_vectors
variance_number = reduce((data - batch_mean) ** 2, "h n d -> h 1 d", "sum")
distributed.all_reduce(variance_number)
batch_variance = variance_number / num_vectors
self.update_with_decay("batch_variance", batch_variance, self.affine_param_batch_decay)

2
lerobot/common/robot_devices/control_configs.py
View File

@@ -66,7 +66,7 @@ class RecordControlConfig(ControlConfig):
private: bool = False
# Add tags to your dataset on the hub.
tags: list[str] | None = None
# Number of subprocesses handling the saving of frames as PNGs. Set to 0 to use threads only;
# Number of subprocesses handling the saving of frames as PNG. Set to 0 to use threads only;
# set to ≥1 to use subprocesses, each using threads to write images. The best number of processes
# and threads depends on your system. We recommend 4 threads per camera with 0 processes.
# If fps is unstable, adjust the thread count. If still unstable, try using 1 or more subprocesses.

4
lerobot/common/robot_devices/motors/dynamixel.py
View File

@@ -242,7 +242,7 @@ class DriveMode(enum.Enum):
class CalibrationMode(enum.Enum):
# Joints with rotational motions are expressed in degrees in nominal range of [-180, 180]
DEGREE = 0
# Joints with linear motions (like gripper of Aloha) are experessed in nominal range of [0, 100]
# Joints with linear motions (like gripper of Aloha) are expressed in nominal range of [0, 100]
LINEAR = 1
@@ -610,7 +610,7 @@ class DynamixelMotorsBus:
# 0-centered resolution range (e.g. [-2048, 2048] for resolution=4096)
values[i] = values[i] / HALF_TURN_DEGREE * (resolution // 2)
# Substract the homing offsets to come back to actual motor range of values
# Subtract the homing offsets to come back to actual motor range of values
# which can be arbitrary.
values[i] -= homing_offset

6
lerobot/common/robot_devices/motors/feetech.py
View File

@@ -221,7 +221,7 @@ class DriveMode(enum.Enum):
class CalibrationMode(enum.Enum):
# Joints with rotational motions are expressed in degrees in nominal range of [-180, 180]
DEGREE = 0
# Joints with linear motions (like gripper of Aloha) are experessed in nominal range of [0, 100]
# Joints with linear motions (like gripper of Aloha) are expressed in nominal range of [0, 100]
LINEAR = 1
@@ -591,7 +591,7 @@ class FeetechMotorsBus:
# 0-centered resolution range (e.g. [-2048, 2048] for resolution=4096)
values[i] = values[i] / HALF_TURN_DEGREE * (resolution // 2)
# Substract the homing offsets to come back to actual motor range of values
# Subtract the homing offsets to come back to actual motor range of values
# which can be arbitrary.
values[i] -= homing_offset
@@ -632,7 +632,7 @@ class FeetechMotorsBus:
track["prev"][idx] = values[i]
continue
# Detect a full rotation occured
# Detect a full rotation occurred
if abs(track["prev"][idx] - values[i]) > 2048:
# Position went below 0 and got reset to 4095
if track["prev"][idx] < values[i]:

4
lerobot/common/robot_devices/robots/dynamixel_calibration.py
View File

@@ -87,7 +87,7 @@ def run_arm_calibration(arm: MotorsBus, robot_type: str, arm_name: str, arm_type
# For instance, if the motor rotates 90 degree, and its value is -90 after applying the homing offset, then we know its rotation direction
# is inverted. However, for the calibration being successful, we need everyone to follow the same target position.
# Sometimes, there is only one possible rotation direction. For instance, if the gripper is closed, there is only one direction which
# corresponds to opening the gripper. When the rotation direction is ambiguous, we arbitrarely rotate clockwise from the point of view
# corresponds to opening the gripper. When the rotation direction is ambiguous, we arbitrarily rotate clockwise from the point of view
# of the previous motor in the kinetic chain.
print("\nMove arm to rotated target position")
print("See: " + URL_TEMPLATE.format(robot=robot_type, arm=arm_type, position="rotated"))
@@ -115,7 +115,7 @@ def run_arm_calibration(arm: MotorsBus, robot_type: str, arm_name: str, arm_type
# TODO(rcadene): make type of joints (DEGREE or LINEAR) configurable from yaml?
if robot_type in ["aloha"] and "gripper" in arm.motor_names:
# Joints with linear motions (like gripper of Aloha) are experessed in nominal range of [0, 100]
# Joints with linear motions (like gripper of Aloha) are expressed in nominal range of [0, 100]
calib_idx = arm.motor_names.index("gripper")
calib_mode[calib_idx] = CalibrationMode.LINEAR.name

2
lerobot/common/robot_devices/robots/feetech_calibration.py
View File

@@ -443,7 +443,7 @@ def run_arm_manual_calibration(arm: MotorsBus, robot_type: str, arm_name: str, a
# For instance, if the motor rotates 90 degree, and its value is -90 after applying the homing offset, then we know its rotation direction
# is inverted. However, for the calibration being successful, we need everyone to follow the same target position.
# Sometimes, there is only one possible rotation direction. For instance, if the gripper is closed, there is only one direction which
# corresponds to opening the gripper. When the rotation direction is ambiguous, we arbitrarely rotate clockwise from the point of view
# corresponds to opening the gripper. When the rotation direction is ambiguous, we arbitrarily rotate clockwise from the point of view
# of the previous motor in the kinetic chain.
print("\nMove arm to rotated target position")
print("See: " + URL_TEMPLATE.format(robot=robot_type, arm=arm_type, position="rotated"))

12
lerobot/common/robot_devices/robots/manipulator.py
View File

@@ -44,7 +44,7 @@ class ManipulatorRobot:
# TODO(rcadene): Implement force feedback
"""This class allows to control any manipulator robot of various number of motors.
Non exaustive list of robots:
Non exhaustive list of robots:
- [Koch v1.0](https://github.com/AlexanderKoch-Koch/low_cost_robot), with and without the wrist-to-elbow expansion, developed
by Alexander Koch from [Tau Robotics](https://tau-robotics.com)
- [Koch v1.1](https://github.com/jess-moss/koch-v1-1) developed by Jess Moss
@@ -55,7 +55,7 @@ class ManipulatorRobot:
robot = ManipulatorRobot(KochRobotConfig())
```
Example of overwritting motors during instantiation:
Example of overwriting motors during instantiation:
```python
# Defines how to communicate with the motors of the leader and follower arms
leader_arms = {
@@ -90,7 +90,7 @@ class ManipulatorRobot:
robot = ManipulatorRobot(robot_config)
```
Example of overwritting cameras during instantiation:
Example of overwriting cameras during instantiation:
```python
# Defines how to communicate with 2 cameras connected to the computer.
# Here, the webcam of the laptop and the phone (connected in USB to the laptop)
@@ -348,7 +348,7 @@ class ManipulatorRobot:
set_operating_mode_(self.follower_arms[name])
# Set better PID values to close the gap between recorded states and actions
# TODO(rcadene): Implement an automatic procedure to set optimial PID values for each motor
# TODO(rcadene): Implement an automatic procedure to set optimal PID values for each motor
self.follower_arms[name].write("Position_P_Gain", 1500, "elbow_flex")
self.follower_arms[name].write("Position_I_Gain", 0, "elbow_flex")
self.follower_arms[name].write("Position_D_Gain", 600, "elbow_flex")
@@ -500,7 +500,7 @@ class ManipulatorRobot:
self.logs[f"read_camera_{name}_dt_s"] = self.cameras[name].logs["delta_timestamp_s"]
self.logs[f"async_read_camera_{name}_dt_s"] = time.perf_counter() - before_camread_t
# Populate output dictionnaries
# Populate output dictionaries
obs_dict, action_dict = {}, {}
obs_dict["observation.state"] = state
action_dict["action"] = action
@@ -540,7 +540,7 @@ class ManipulatorRobot:
self.logs[f"read_camera_{name}_dt_s"] = self.cameras[name].logs["delta_timestamp_s"]
self.logs[f"async_read_camera_{name}_dt_s"] = time.perf_counter() - before_camread_t
# Populate output dictionnaries and format to pytorch
# Populate output dictionaries and format to pytorch
obs_dict = {}
obs_dict["observation.state"] = state
for name in self.cameras:

4
lerobot/common/robot_devices/robots/stretch.py
View File

@@ -108,7 +108,7 @@ class StretchRobot(StretchAPI):
self.logs[f"read_camera_{name}_dt_s"] = self.cameras[name].logs["delta_timestamp_s"]
self.logs[f"async_read_camera_{name}_dt_s"] = time.perf_counter() - before_camread_t
# Populate output dictionnaries
# Populate output dictionaries
obs_dict, action_dict = {}, {}
obs_dict["observation.state"] = state
action_dict["action"] = action
@@ -153,7 +153,7 @@ class StretchRobot(StretchAPI):
self.logs[f"read_camera_{name}_dt_s"] = self.cameras[name].logs["delta_timestamp_s"]
self.logs[f"async_read_camera_{name}_dt_s"] = time.perf_counter() - before_camread_t
# Populate output dictionnaries
# Populate output dictionaries
obs_dict = {}
obs_dict["observation.state"] = state
for name in self.cameras: