Predict a chunk of actions given environment observations.

Return a chunk of 20 dummy actions.

Builds a 'Read' status packet. Args: scs_id (int): ID of the servo responding. value (int): Desired value to be retur

Ensure bimanual PiPER configs are not mixed between PiPER and PiPER-X variants.

Save annotations to LeRobot dataset parquet format.

This function creates the following directory structure: 005000/ # training step at checkpoint ├── pretrained_model/ │ ├── config.jso

Set current position as zero for selected motors.

Start the robot client and connect to the policy server

Builds a "Sync_Read" broadcast instruction. The parameters for Sync Read are: param[0] = start_address par

Converts all transitions in this ReplayBuffer into a single LeRobotDataset object.

Applies feature transformations from all steps sequentially. This method propagates a feature description dictionary through each step's

Transforms feature keys from the Gym standard to the LeRobot standard. This method standardizes the feature dictionary by renaming k

Removes a processor step from the registry. Args: name: The name of the step to unregister.

Initialize the publisher (no-op for ZMQ).

Add a legend outside the plot area on the right side. Args: fig: Matplotlib figure to add legend to axs: Array of axe

Enqueue an observation if it must go through processing, otherwise skip it. Observations not in queue are never run through the policy network

Get current robot observation including joint positions and camera images.

Check if config follows DataProcessorPipeline format. This method validates the processor configuration structure: **Required Struct

Check if the observation is valid to be processed by the policy

If last dim is REAL_DIM (12), pad zeros to reach dim_action (20).

Helper method to prepare 4D attention masks for transformer.

Rescale axes to show all data with proper margins. Args: axes: Array of matplotlib axes to rescale

Helper to send simple 8-byte commands (Enable, Disable, Zero).

Savitzky-Golay smoothing for 1-D array. Use window=1 to disable.

Validate that a config was loaded and is a valid processor config. This method validates processor config format with intelligent migration d

Validate that all provided overrides were used. This method ensures user overrides are valid to catch typos and configuration errors:

Convert rotation to 3x3 rotation matrix. Returns: 3x3 rotation matrix

Test that we can iterate through the entire dataset without errors.

Helper to assert two transitions are equal.

Reads the latest available frame data (color) asynchronously. This method retrieves the most recent color frame captured by the back

This method forwards all its arguments to LlamaTokenizerFast's [`~PreTrainedTokenizer.batch_decode`]. Please refer to the docstring o

Constructs a model-ready observation tensor dict from a raw observation. This utility function orchestrates the process of converting a raw,

Default: raise if the type is not recognized.

Compute the sampling weights for the online training dataloader in train.py. Args: offline_dataset: The LeRobotDataset used for offline p

Concatenate multiple video files into a single video file using pyav. This function takes a list of video input file paths and concatenates

Establishes ZMQ sockets with the remote mobile robot

Connects to the OpenCV camera specified in the configuration. Initializes the OpenCV VideoCapture object, sets desired camera proper

Create a `DatasetCard` for a LeRobot dataset. Keyword arguments are used to replace values in the card template. Note: If specified, `license

Disable torque on selected motors.

Loads a pipeline from a local directory, single file, or Hugging Face Hub repository. This method implements a simplified loading pipeline wi

Extract and optionally cache image features from observations. This function processes image observations through the vision encoder once and

Returns the serializable configuration of the processor. Returns: A dictionary containing the device and float_dtype set

Creates an infinite iterator that yields batches of transitions. Will automatically restart when internal iterator is exhausted.

Get the version of a package if it exists, otherwise return 'N/A'.

Returns the checkpoint sub-directory corresponding to the step number.

Linearly interpolate values to fill in gaps (robust to NaNs / edge cases).

Check if a feature key represents an image feature. Args: key: The feature key to check Returns: True if the key represents

hf_dataset contains all the observations, states, actions, rewards, etc.

Load optimizer state from disk. Args: optimizer: Either a single optimizer or a dictionary of optimizers. save_dir: Directory to

Generates the optimizer and scheduler based on configs. Args: cfg (TrainPipelineConfig): The training config that contains optimizer and

Recursively move all tensors in a (potentially) nested dict/list/tuple structure to the CPU.

Return the 95th percentile latency or None if empty.

Samples an action from the learned policy. The policy can also have added (truncated) Gaussian noise injected for encouraging exploration whe

Builds a 'Ping' status packet. https://emanual.robotis.com/docs/en/dxl/protocol2/#ping-0x01 Args: dxl_id (int):

Process a single episode annotation.