Exception raised when the robot device is not connected.

Normalizes data (e.g. "observation.image") for more stable and faster convergence during training.

ResNet style 1D convolutional block with FiLM modulation for conditioning.

Provides a type for a dataset containing video frames. Example: ```python data_dict = [{"image": {"path": "videos/episode_0.mp4", "

Exception raised when the robot device is already connected.

Similar to `Normalize` but unnormalizes output data (e.g. `{"action": torch.randn(b,c)}`) in their original range used by the environment.

Conv1d --> GroupNorm --> Mish

Convenience module for running multiple encoder layers, maybe followed by normalization.

Configuration class for DiffusionPolicy. Defaults are configured for training with PushT providing proprioceptive and single camera observations.

Action Chunking Transformer: The underlying neural network for ACTPolicy. Note: In this code we use the terms `vae_encoder`, 'encoder', `decoder`

Configuration class for the Action Chunking Transformers policy. Defaults are configured for training on bimanual Aloha tasks like "insertion" or

2D sinusoidal positional embeddings similar to what's presented in Attention Is All You Need. The variation is that the position indices are norm

A 1D convolutional UNet with FiLM modulation for conditioning. Note: this removes local conditioning as compared to the original diffusion policy

Diffusion Policy as per "Diffusion Policy: Visuomotor Policy Learning via Action Diffusion" (paper: https://arxiv.org/abs/2303.04137, code: h

Encoder an RGB image into a 1D feature vector. Includes the ability to normalize and crop the image first.

1D sinusoidal positional embeddings as in Attention is All You Need.

From https://github.com/notmahi/miniBET/blob/main/behavior_transformer/bet.py

Original comments: Full definition of a GPT Language Model, all of it in this single file. References: 1) the official GPT-2 TensorFl

JPEG 2000 codec for numcodecs.

JPEG XL codec for numcodecs.

Example of usage: ```python KochRobotConfig() ```

Primary logger object. Logs either locally or using wandb. The logger creates the following directory structure: provided_log_dir ├── .h

A dataset consisting of multiple underlying `LeRobotDataset`s. The underlying `LeRobotDataset`s are effectively concatenated, and this class adop

FIFO data buffer for the online training loop in train.py. Follows the protocol of LeRobotDataset as much as is required to have it be used by th

The OpenCVCamera class allows to efficiently record images from cameras. It relies on opencv2 to communicate with the cameras. Most cameras a

Example of tested options for Intel Real Sense D405: ```python OpenCVCameraConfig(30, 640, 480) OpenCVCameraConfig(60, 640, 480)

Apply a random subset of N transformations from a list of transformations. Args: transforms: list of transformations. p: represen

Residual VQ is composed of multiple VectorQuantize layers. Follows Algorithm 1. in https://arxiv.org/pdf/2107.03312.pdf "Residual Ve

Randomly change the sharpness of an image or video. Similar to a v2.RandomAdjustSharpness with p=1 and a sharpness_factor sampled randomly. W

Spatial Soft Argmax operation described in "Deep Spatial Autoencoders for Visuomotor Learning" by Finn et al. (https://arxiv.org/pdf/1509.061

Spatial Soft Argmax operation described in "Deep Spatial Autoencoders for Visuomotor Learning" by Finn et al. (https://arxiv.org/pdf/1509.061

Configuration class for TDMPCPolicy. Defaults are configured for training with xarm_lift_medium_replay providing proprioceptive and single ca

Encode image and/or state vector observations.

Task-Oriented Latent Dynamics (TOLD) model used in TD-MPC.

Measures execution time using a context manager or decorator. This class supports both context manager and decorator usage, and is thread-sa

A class representing a UR robot.

Configuration class for VQ-BeT. Defaults are configured for training with PushT providing proprioceptive and single camera observations. The

VQ-BeT: The underlying neural network for VQ-BeT Note: In this code we use the terms `rgb_encoder`, 'policy', `action_head`. The meanings are as

Encode an RGB image into a 1D feature vector. Includes the ability to normalize and crop the image first. Same with DiffusionRgbEncoder from

Example of usage: ```python gello_ur5RobotConfig() ```

Action Chunking Transformer Policy as per Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware (paper: https://arxiv.org/abs/23

This class allows to control any Koch robot of various number of motors. A few versions are available: - [Koch v1.0](https://github.com/Alexa

The required interface for implementing a policy. We also expect all policies to subclass torch.nn.Module and PyTorchModelHubMixin.

Zarr-based temporal datastructure. Assumes first dimension to be time. Only chunk in time dimension.

Implementation of TD-MPC learning + inference. Please note several warnings for this policy. - Evaluation of pretrained weights created w

VQ-BeT Policy as per "Behavior Generation with Latent Actions"