Return an activation function given a string.

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

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

Check whether the python process was launched through slurm

Detects if python is running without a monitor.

Predict the next state's latent representation and the reward given a current latent and action. Args: z: (*, latent_dim) tensor

Log message with level WARNING.

Args: cfg: A Hydra config as per the LeRobot config scheme. split: Select the data subset used to create an instance of LeRobotDa

Convert environment observation to LeRobot format observation. Args: observation: Dictionary of observation batches from a Gym vector envi

Squeeze dim only if it's not the last dim. Image dim expected to be *, H, W, C

Read a frame from the camera returned in the format (height, width, channels) (e.g. (640, 480, 3)), contrarily to the pytorch format which is

This function ensures that a neural network trained on data collected on a given robot can work on another robot. For instance before calibration,

Set the random state for `random`, `numpy`, and `torch`. Args: random_state_dict: A dictionary of the form returned by `get_global_random

Command the leader robot to a given state. Args: joint_state (np.ndarray): The state to command the leader robot to.

Update the target model's parameters with an EMA step.

Returns a dictionary of items for logging.

Args: shapes (dict): A dictionary where keys are input modalities (e.g. "observation.image") and values are their sha

Convenience function for `eval_policy(return_episode_data=True)` Compiles all the rollout data into a Hugging Face dataset. Similar logic is

Initialize model weights. Orthogonal initialization for all linear and convolutional layers' weights (apart from final layers of rewa

Makes the data spec for np.memmap.

Make a numpy memmap with checks on available disk space first. Expected kwargs are: "filename", "dtype" (must by np.dtype), "mode" and "shape"

Factory for noise scheduler instances of the requested type. All kwargs are passed to the scheduler.

Returns path1 relative to path2.

Args: root_module: The module for which the submodules need to be replaced predicate: Takes a module as an argument and must retu

Args: root_module: The module for which the submodules need to be replaced predicate: Takes a module as an argument and must retu

Xavier-uniform initialization of the transformer parameters as in the original code.

Add new data to the buffer, which could potentially mean shifting old data out. The new data should contain all the frames (in order) of any

Aggregate stats of multiple LeRobot datasets into one set of stats without recomputing from scratch. The final stats will have the union of all d

Convert from unsigned int32 joint position range [0, 2**32[ to the universal float32 nominal degree range ]-180.0, 180.0[ with a "zero positio

This function expects `batch` to have (at least): { "observation.state": (B, n_obs_steps, state_dim) "observ

This long function is unfortunately doing something very simple and is being very defensive: We are separating out all parameters of

Copy a on-disk zarr to in-memory compressed. Recommended

Load to memory.

Create a branch on a existing Hugging Face repo. Delete the branch if it already exists before creating it.

1D sinusoidal positional embeddings as in Attention is All You Need. Args: num_positions: Number of token positions required. Returns

Run this to get basic system info to help for tracking issues & bugs.