TODO(allenzren): does this dummy env allow camera obs for other envs besides robomimic?

Convert Euler Angles to Rotation Matrix. See rotation.py for notes

Use target actor

Inputs: action: (B, Ta, Da) - Action trajectories time: (B,) or float - Flow time d: (B,) or float - Step

Differentiable forward pass used in actor training.

Generate rectified flow trajectory xt = t * x1 + (1 - t) * x0. Args: x1: Target data tensor of shape (batch_size, horizon_steps,

Download processed train.npz and normalization.npz to the dataset paths specified in cfg.

Use torchdiffeq samples: (B x Ta x Da)

Create a function to compute the unbiased log-likelihood estimate of a given data point. Args: sde: A `sde_lib.SDE` object that represents

inputs: x_chain: torch.Tensor of shape `[batchsize, self.inference_steps+1, self.horizon_steps, self.action_dim]`

Calculating the logprobs of random samples of denoised chains. Args: cond: dict with key state/rgb; more recent obs at t

get num_ind random indices from a set of size sz (used for getting critic targets)

get num_ind random indices from a set of size sz (used for getting critic targets)

input: cond: dict, which contains conda["state"]: torch.Tensor(self.n_envs, self.n_cond_steps, self.obs_dim) output:

Wraps `score_fn` so that the model output corresponds to a real time-dependent score function. Args: sde: An `sde_lib.SDE` object that repr

for evaluation in sim

loads model from disk

loads model and ema from disk

Iterates through all Python files in the specified directories and adds the MIT License.

Main function to load configuration, process evaluation statistics, and generate 3D surface plots comparing pretrained and finetuned models.

makes indices for sampling from dataset; each index maps to a datapoint, also save the number of steps before it within the same traj

Convert Rotation Matrix to Quaternion. See rotation.py for notes

Puts angles in [-pi, pi] range.

If predicting epsilon: E_{t, x0, ε} [||ε - ε_θ(√α̅ₜx0 + √(1-α̅ₜ)ε, t)||² Args: x_start: (batch_size, horizon_ste