A factory for creating pl.DataModule. Valid module class names: "L5MixedDataModule", "L5RasterizedDataModule" Args: cls_name (st

noise_hist: [B, M, T_hist, 2] TBD: get history diffuser norm info to apply

Expand all tensors in nested dictionary or list or tuple at a single dimension @dim by @size. Args: x (dict or list or tuple): a

Expand a tensor at a single dimension @dim by @size Args: x (torch.Tensor): input tensor size (int): size to expand

Flatten a nested dict or list to a list. For example, given a dict { a: 1 b: { c: 2 } c: 3

Flatten a tensor in all dimensions from @begin_axis onwards. Args: x (torch.Tensor): tensor to flatten begin_axis (int): whi

Get a batch of prior parameters. Args: inputs (torch.Tensor): (Optional) A feature vector for priors that are input-cond

Get a batch of prior parameters. Args: inputs (torch.Tensor): (Optional) A feature vector for priors that are input-cond

Create a SimulationDataset by picking indices from the provided dataset :param dataset: the EgoDataset :param scene_indices: scenes f

Apply @gather_along_dim_with_dim_single to all tensors in a nested dictionary or list or tuple. Args: x (dict or list or tuple):

- pos : B x 2 - yaw : B x 1 - lw : B x 2

This version generates ROI for all agents only at most recent time step unless specified otherwise

mask out area near the existing samples to boost diversity Args: orig_loc (torch.tensor): original sample location, 1 for sample, 0 for

Get episodic experience in the form of datasets

Args: avails (torch.Tensor): target availabilities [B, (A), T] Returns: last_indices (torch.Tensor): index of the last avail

sets loss coefficients for trajectory action_weight : float coefficient on first action loss d

sets loss coefficients for trajectory action_weight : float coefficient on first action loss d

Extract the "mean" of a prior distribution (not supported by all priors) Args: prior_params (torch.Tensor): a batch of p

get the neighbor history relative states (only need once per data_batch). We do this because fields like all_other_agents_history_positions i

Get left/right lane of current lane - param current_lane: lane object - param neighbor_lanes: (S_seg, S_point, 3) - param agent_from_

Extract state and(or) action from the data_batch from data_batch Note: Support both agent-centric and scene-centric (extr

Extract state and(or) action from the data_batch from data_batch Input: data_batch: dict Output: x: (

Reward that incentivizes progress.

Returns metrics objects to measure success of each guidance.

Rollout an environment. Args: env (BaseEnv): a base simulation environment (gym-like) policy (RolloutWrapper): a policy that

Sets a global target pos and time using GT.

Sets a local target pos and time using GT.

Initializes this loss to be used repeatedly only for the given scenes/agents in the example_batch. e.g. this function could use the e

Integrate the state forward with initial state x0, action u Args: x0 (Torch.tensor): state tensor of size [B,Num_agent,1,

Compute kl loss between the prior and the posterior distributions. Args: posterior_params (dict): a batch of distributio

Loads a boolean mask of the agent availability stored into the zarr. Performs some sanity check against cfg. Returns: a boolean mask

Compute the log(mean(exp(x), dim)) in a numerically stable manner. Adapted from CS 236 at Stanford. Args: x (torch.Tensor): a ten