An update method that is to be called after a training optimization step. Implements an additional updates the model parameters may need (for

Calculate episode data index for the provided HuggingFace Dataset. Relies on episode_index column of hf_dataset. Parameters: - hf_datase

This function convert the degree range to the step range for indicating motors rotation. It assums a motor achieves a full rotation by going from

More info on ffmpeg arguments tuning on `benchmark/video/README.md`

Get a module's device by checking one of its parameters. Note: assumes that all parameters have the same device

Given the log directory, get the sub-directory in which the last checkpoint will be saved.

Initialize a Hydra config given only the path to the relevant config file. For config resolution, it is assumed that the config file's parent is

Predict state-action value for all of the learned Q functions. Args: z: (*, latent_dim) tensor for the current state's latent rep

GPT model gets hyperparameters from a config object. Please refer configuration_vqbet.py for more details.

Returns the default ffmpeg encoding parameters used by `encode_video_frames`.

Given a string, return a torch.device with checks on whether the device is available.

Make an instance of a policy class. Args: hydra_cfg: A parsed Hydra configuration (see scripts). If `pretrained_policy_name_or_path` is

Set seed for reproducibility.

Encodes an observation into its latent representation.

Copied from https://github.com/huggingface/transformers/blob/main/src/transformers/utils/import_utils.py Check if the package spec exists and grab

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

To be called whenever the environment is reset. Does things like clearing caches.

Return one action to run in the environment (potentially in batch mode). When the model uses a history of observations, or outputs a sequence

Predict state value (V). Args: z: (*, latent_dim) tensor for the current state's latent representation. Returns:

Loads frames associated to the requested timestamps of a video The backend can be either "pyav" (default) or "video_reader". "video_reader" r

Disconnect from the robot.

Args: env: The batch of environments. policy: The policy. n_episodes: The number of episodes to evaluate. max_epi

Get the random state for `random`, `numpy`, and `torch`.

Get the current state of the leader robot. Returns: T: The current state of the leader robot.

Makes a gym vector environment according to the evaluation config. n_envs can be used to override eval.batch_size in the configuration. Must be a

Resolves delta_timestamps config key (in-place) by using `eval`. Doesn't do anything if delta_timestamps is not specified or has already been res

Encode image features and concatenate them all together along with the state vector.

Return a group name for logging. Optionally returns group name as list.

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

Compute mean/std and min/max statistics of all data keys in a LeRobotDataset.

Create buffers per modality (e.g. "observation.image", "action") containing their mean, std, min, max statistics. Args: (see Normalize a

The equivalent of itertools.cycle, but safe for Pytorch dataloaders. See https://github.com/pytorch/pytorch/issues/23900 for information on why i

Helper to temporarily flatten extra dims at the start of the image tensor. Args: fn: Callable that the image tensor will be passed to. It

Create a LeRobot Dataset from existing data and attributes instead of loading from the filesystem. It is especially useful when converting ra