Get the value of a specific key from the tensorboard Args: split (str): The split to get the value from. step

image from the simulator is flipped. so we need to flip it back.

Shared installation logic for all custom install commands. Handles: - Installing pre-commit hooks (if available) - Installing option

Shared installation logic for all custom install commands. Handles: - Installing pre-commit hooks (if available) - Installing option

All the operations that happen in transforming a sample from indiviusal dataset format to Image unifier format. Currently, it just loads

Remove keys from the sample that are not needed.

Saves all information required for resuming training in the experiment folder.

Check if stats exist in cache_dir and if they do, return them. If they don't exist, return None.

Get a yacs CfgNode object with default values for my_project.

Returns dataloader based on the config and split :param get_dataset: whether to return the dataset or the dataloader

Calculates statistics for specified keys over an entire dataset by iteration. This function assumes input data for keys has ndim >= 1 (not s

Given task suite name or task name, return the tasks to evaluate on. Args: task_suite_name (_type_, optional): _description_. Defaul

This is a helper function that is used to get the final le_cam_list and rv_cam_list used in the dataset. It checks if the camera names are va

Get the metadata for a LeRobot dataset :param repo_id: (str) Repository ID from huggingface to load the dataset :return: (dict) Metadata

Returns model based on the config

Returns optimizer and learning rate scheduler based on the config :param cfg: config object :param model: model to optimize :param nu

Loads a pretrained model from a given path. :param model_path: path to the pretrained model :param device: device to load the model on, s

Does the unification from roboverse output to SO100 data

Convert a sample from LeRobot dataset to a sample in RV dataset. Should use the exact same arguments as LeRobotRV.__init__. outputs in Numpy.

Loads a pretrained model from a given path. :param model: model to load :param model_path: path to the pretrained model :param cfg: c

Loads a pretrained model from a given path. :param model: model to load :param optimizer: optimizer to load :param lr_sched: learning

Call the /predict_base64 endpoint to get full action sequence. Args: images: List of RGB images as numpy arrays (H, W, 3) st

Call the /predict_base64_stream endpoint to get actions one at a time. Args: images: List of RGB images as numpy arrays (H, W, 3)

Remove keys from a cfg.

Convert a numpy RGB image array to base64 string.

Set the dataset stats for the model :param dataset_stats: dict, the dataset stats

Fix episode_data_index mapping when episodes are filtered in LeRobotDataset. Uses the same approach as PR #1062: https://github.com/h

Initializes and returns the LIBERO environment, along with the task description.

Given task name, and task suite names, return the task, init states, max steps, and task description. Args: task_name (_type_): _des

Copied from robosuite: https://github.com/ARISE-Initiative/robosuite/blob/eafb81f54ffc104f905ee48a16bb15f059176ad3/robosuite/utils/transform_util

Check if two configs are similar. It ignore the keys that do not affect the stats.

Edit CMakeLists.txt to set cmake_minimum_required to version 3.5

Writes results to results.json file in the following format: {"success": X, "failure": X, "task_suite_name": { "success": X, "fa

Runs 1 episode of the task given the model and environment. Args: env (_type_): _description_ model (_type_): _description_

Example of how to use the API with dummy data.

Convert the data into the format required by the model :param system_message: str, the system message :param image: list of PIL images, t

Get the action from the text action. :param action_txt: List[str], the text action. :param instruction: str, the instruction

Get a yacs CfgNode object with default values for my_project.

Get a dictionary of datasets from the config.

Get the images for the given inputs :param bs: int, the batch size :param pc: torch.Tensor, the point cloud :param rg

Constructs the input for the model using the batched data. :param cfg: config object :param data_batch: contains the batched data provide

Get the installed LeRobot version string.

Get the Qwen inputs for the given inputs :param bs: int, the batch size :param imgs: list of list of PIL images :para

Get the text action from the actions. :param actions: torch.Tensor, the actions to convert to text. :param instruction: str,

Get the unified dataset based on the config file.

Get the stats of the unifier. :param dataset: The dataset to get the stats of. :param cfg: The config to get the stats of. :param sta

Given task name and optionally the task suite name that it is in, initialize the libero environment. Args: verse_config (_type_): _d

Patch LIBERO's benchmark/__init__.py to add weights_only=False to torch.load Issue: PyTorch 2.6+ changed torch.load default from weights_onl

Prints results in a neatly aligned tabular format with dynamic column widths and an OVERALL column. The OVERALL column displays both the aver

Print model statistics including parameter counts.

Orders eval names with model_N.pth_<...> (ascending N) first, others go last.

Tile a images into a single image :param images: Tensor of shape (bs, H, W, 3) or list of tensors of shape (H, W, 3) :return:

Training for one epoch

Does the unification from SO100 data to roboverse data

:param qwen_model_id: str, the id of the qwen model to use :param action_type: str, the type of action to use, either ORIGINAL or EE

:param history: (int) Number of history frames :param horizon: (int) Number of future action :param repo_id: (str) Repository