Generate a prediction based on latent prior (instead of sample) and condition inputs Args: condition_inputs (dict, torch

scale the trajectory from original scale to standard normal distribution Note: Support both agent-centric and scene-centr

Given feature layers and proposals return bilinear interpolated points in feature layer Args: features (torch.Tensor): Tensor of shap

- scene_idx : which scene index this guidance is used for (within each batch, i.e. the LOCAL scene index) - params : param dict that

Convert the input (log-replayed) and output (simulated) trajectories into world space. Apply availability on the log-replayed one :param in_o

Calculate collision loss among predicted edges along a batch of trajectories Args: pred_edges (dict): A dict that maps collision type

Compute KL Divergence loss Args: outputs (dict): outputs of the self.forward() call Returns: a dict

U-Net forward by concatenating input feature (x1) with mirroring encoding feature maps channel-wise (x2) Args: x1 (torch.

Return the minimum number of frames between the scenes :return: the minimum number of frames

Log probability of tensor x under diagonal multivariate normal with mean m and variance v. The last dimension of the tensors is treated a

if self.predict_epsilon, model output is (scaled) noise; otherwise, model predicts x0 directly

if self.predict_epsilon, model output is (scaled) noise; otherwise, model predicts x0 directly

Reshape selected dimensions in a tensor to a target dimension. Args: x (torch.Tensor): tensor to reshape begin_axis (int): b

- traj: B x T x D

invoked 1.at initialization 2.by GlobalStopSignLoss forward to reset stop sign pos at each rollout step

- yaw : (B) - tf_mat : (B, 3, 3) matrix to transform yaw by

Returns true if there list contains some valid guidance that needs to be applied. Does not check to make sure each guidance dict is structure

put the lists of ROI align result into embedding tensor with the help of index

Aggregated per-step loss between gt and predicted trajectories Args: predictions (torch.Tensor): predicted trajectory [B, M, (A), T,

d_edge: applies only when neigh_hist_embed_method is 'edge_interaction' k_attr: number of attributes of history trajectories

A dummy wrapper that add one dimension to each field.

Compute whether the given positions are out of drivable region Args: positions (torch.Tensor): a position (x, y) in rasterized frame

-pos: [B*N, T, 1, 2] -in_stop_box: [B*N, T, 1]

preds: dict of predictions from model, preds["positions"] (M, N, T, 2) or (B, N, M, T, 2)

preds: dict of predictions from model, preds["positions"] (M, N, T, 2) or (B, N, M, T, 2)

http://nbviewer.ipython.org/github/dpsanders/matplotlib-examples/blob/master/colorline.ipynb http://matplotlib.org/examples/pylab_examples/mu

Creates a video from a format for frame e.g. 'data_out/frame%04d.png'. Saves in out_path.

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

Queries the sim scene for current state of all agents returns: - curr_agent_state: (N x 8) [pos, vel, acc, heading_angle]

Query projected lane points in agent coordinate by passing in predicted trajectories in agent coordinate. - param pos_pred (B, N, T, 2) -

If using the model as a trajectory predictor (generating trajs for non-ego agents)

return stl robustness for the passed signal

Returns state dictionary from pl_module. Override if you want filter some parameters and/or buffers out. For example, in pl_module has metrics

This function is called when disable_control_on_stationary is not False.

This function provides weight and bias initializations for linear layers.

- x : (B, M, T, D) - map_grid_feat : (B, M, C, H, W) - raster_from_agent: (B, M, 3, 3) - output feats_out : (B, M, T