Iterate over all contact points of the actor.

Joins all dimensions between dimensions (@begin_axis, @end_axis) into a flat dimension, for all tensors in nested dictionary or list or tuple

Joins all dimensions between dimensions (@begin_axis, @end_axis) into a flat dimension, for all tensors in nested dictionary or list or tuple

Loads and preprocesses all episodes. Resizes all images in one episode before loading the next, to reduce memory usage.

Loads a checkpoint for a given module. Args: module_name (str): Name of model component to load checkpoint for. path (str):

Log metrics to Weights & Biases. Args: metrics (dict): Dictionary of metrics to log prefix (str): Prefix for metric names

Adapted from big_vision. Tokens can attend to valid inputs tokens which have a cumulative mask_ar smaller or equal to theirs. This way `mask_

Adapted from big_vision. Tokens can attend to valid inputs tokens which have a cumulative mask_ar smaller or equal to theirs. This way `mask_

Converts rotation matrices to 6D rotation representation by Zhou et al. [1] by dropping the last row. Note that 6D representation is not uniq

Forward pass of DiT. x: (N, T, input_dim) t: (N,) tensor of diffusion timesteps global_cond: (N, n_obs_steps, D) tens

x: (B,T,input_dim) timestep: (B,) or int, diffusion step global_cond: (B,global_cond_dim) output: (B,T,input_dim)

v: (..., N)

Serialize Python data (including numpy arrays) to JSON string

Function to compute output shape from inputs to this module. Args: input_shape (iterable of int): shape of input. Does n

Pad input tensor or array @seq in the time dimension (dimension 1). Args: seq (np.ndarray or torch.Tensor): sequence to be padded

Pad input tensor or array @seq in the time dimension (dimension 1). Args: seq (np.ndarray or torch.Tensor): sequence to be padded

[Modify] Parse a hdf5 file to generate a state trajectory. Args: file_path (str): the path to the hdf5 file Returns:

Interpolative planning with screw motion. Will not avoid collision and will fail if the path contains collision.

计算cos和sin的值，cos值在实部，sin值在虚部，类似于 cosx+j*sinx :param dim: q,k,v的最后一维，一般为emb_dim/head_num :param end: 句长length :param constant： 这里指10000

obs_dict: str: B,To,* return: B,Ta,Da

Print the number of parameters in each part of the model.

Randomly swaps the two exterior images (for training with single exterior input).

Read the dirty bit from the chunk directory.

Read the dirty bit from the chunk directory.

Re-indexes a batch in a way that is conducive to DistributedSampler + grouping by seqlen per rank.

Relabels actions to use reached proprioceptive state; discards last timestep (no-action).

This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch, num_key_value_heads, seqlen, he

Rotates half the hidden dims of the input.

Converts 6D rotation representation by Zhou et al. [1] to rotation matrix using Gram--Schmidt orthogonalization per Section B of [1]. Arg

Runs the runtime loop continuously until stop() is called or the environment is done.

Runs the runtime loop continuously until stop() is called or the environment is done.

Compute model forward pass and metrics for both training and validation. Args: vla (OpenVLAForActionPrediction): Vision-language-act

Saves a `dataset_statistics.json` file.

Save a sample to the chunk directory.

Selects the best resolution from a list of possible resolutions based on the original size. Args: original_size (tuple): The origina

Sets the number of input images for the vision backbone. Args: num_images_in_input: Number of images to expect in the in