r"""Take an action for each sub-environments. Parameters ---------- actions : element of :attr:`action_space` Bat

loads model from disk

train_prt_str_additional: str, additional information in training that will be printed to log console that is not included in train_prt_str_b

Sample trajectories using the learned velocity field. Args: cond: Dictionary containing 'state' tensor of shape (batch_size, cond

Compute the MSE loss between predicted and target velocities. Args: xt: Corrupted data tensor of shape (batch_size, horizon_steps

saves model to disk; no ema

r"""Close all sub-environments and release resources. It also closes all the existing image viewers, then calls :meth:`close_extras` and set

Convert Quaternion to Euler Angles. See rotation.py for notes

Abstraction from os.path.join() Builds absolute paths from relative path strings with SIM_FRAMEWORK/ as root. If args already contains an

**Args**: action: (B, Ta, Da) time: (B,) or int, diffusion step cond: dict with key state/rgb; more recen

Set the random seed in all sub-environments. Parameters ---------- seed : list of int, or int, optional Random se

obs_venv: dict containing numpy.ndarray

Inputs: action: (B, Ta, Da) - Current action trajectory time: (B,) - Current noise level t cond: (B, Do)

examples: get_schedule('constant', level=0.2)) get_schedule('linear', max=0.2, hold_steps=100, anneal_steps=300, min=0.1))

r"""Reset all sub-environments and return a batch of initial observations. Returns ------- element of :attr:`observation_spac

Generate actions for a batch of observations using the actor with random noise. Args: cond (Dict[str, Tensor]): Dictionary contai

inputs: cond: dict, contatinin... 'state': obs. observation in robotics. torch.Tensor(batchsize, cond_steps, obs_

Calls the MLP to compute the mean, scale, and logits of the GMM. Returns the torch.Distribution object.

x: (B, Ta, act_dim) time: (B,) or int, diffusion step cond: dict with key state/rgb; more recent obs at the end s

normalize self.reward_trajs

cond: dict with key state/rgb; more recent obs at the end state: (B, To, Do) or (B, num_feature) from ViT encoder

Create a vectorized environment from multiple copies of an environment, from its id. Parameters ---------- env_name : str The

q(xₜ | x₀) = 𝒩(xₜ; √ α̅ₜ x₀, (1-α̅ₜ)I) xₜ = √ α̅ₜ xₒ + √ (1-α̅ₜ) ε

Reset one environment with options.

Output (n_steps,) + obs_shape

DDPM parameters

Remove all __pycache__ directories and .pyc files in the given directory and its subfolders. Args: directory (str): The root dir

Extract the first number from a label string.

x: (B, Ta, Da) time: (B,) or int, diffusion step cond: dict with key state/rgb; more recent obs at the end state:

use both pre-trained and online policies

use both pre-trained and online policies

Not the best design

create noise std for logrporbability calcualion. generate a tensor `self.logprob_noise_levels` of shape `[1, self.denoising_steps,

Apply padding

bugfix:

Get the connection object needed to communicate with the wanted envs that are in subprocesses.