Estimates initial curvature (curvature_0) and curvature rate ({curvature_rate_0, ...}) using least squares with curvature rate regularization

Estimates initial velocity (v_0) and acceleration ({a_0, ...}) using least squares with jerk penalty regularization. :param xy_displacements:

predictions: (n_agent, 80, 2 or 3)

predictions: (n_agent, 80, 2 or 3)

note that rear axle velocity and acceleration are in ego local frame, and need to be transformed to the global frame.

note that rear axle velocity and acceleration are in ego local frame, and need to be transformed to the global frame.

Returns on-route lanes and heading errors where ego-vehicle intersects. :param ego_state: state of ego-vehicle :return: tuple

Returns the most suitable starting lane, in ego's vicinity. :param ego_state: state of ego-vehicle :return: lane object (on-r

Returns position and heading displacements given a pose trajectory. :param poses: <np.ndarray: num_poses, 3> A trajectory of poses (x, y, hea

This lateral controller determines a steering_rate input to minimize lateral errors at a lookahead time. It requires a velocity seque

Logs the artifacts from a training/validation/test step. :param loss: scalar loss value :type objectives: [type] :pa

Logs the artifacts from a training/validation/test step. :param loss: scalar loss value :type objectives: [type] :pa

This longitudinal controller determines an acceleration input to minimize velocity error at a lookahead time. :param initial_velocity

Returns a banded difference matrix with specified number_rows. When applied to a vector [x_1, ..., x_N], it returns [x_2 - x_1, ..., x_N - x_

Returns an array of heading angles equal mod 2 pi to the input heading angles, and such that the difference between successive output angles

Given y, it solves argmin_z |y-z|_2 st \sum z = 1 , 1 >= z_i >= 0 for all i

This function uses LQR to find an optimal input to minimize tracking error in one step of dynamics. The dynamics are next_state = A @

This function uses LQR to find an optimal input to minimize tracking error in one step of dynamics. The dynamics are next_state = A @

Apply proportional controller when at near-stop conditions. :param initial_velocity: [m/s] The current velocity of ego. :para

candidate_trajectories: (n_ref, n_mode, 80, 3) probability: (n_ref, n_mode)

candidate_trajectories: (n_ref, n_mode, 80, 3) probability: (n_ref, n_mode)

This function applies some first order control delay/a low pass filter to acceleration/steering. :param state: Ego state :pa

params: center: [bs, N, (x, y)] heading: [bs, N] width: [bs, N] length: [bs, N] retur

Builds the lightning trainer from the config. :param cfg: omegaconf dictionary :return: built object.

Build the lightning datamodule from the config. :param cfg: Omegaconf dictionary. :param model: NN model used for training. :param wo

Builds the lightning module from the config. :param cfg: omegaconf dictionary :param torch_module_wrapper: NN model used for training

Cleanup lingering s3 artifacts that are written locally. This happens because some minor write-to-s3 functionality isn't yet implemented.

Initiliazes WeightedSampler object with sampling weights for each scenario_type and returns it. :param scenario_dataset: ScenarioDataset obje

Accumulates all within-bound comfortability metrics :param states: array representation of ego state values :param time_point_s: time ste

Converts an ego state into an array representation (drops time-stamps and vehicle parameters) :param ego_state: ego state class :return:

Given a list of vectors (vecs), this method finds the minimum norm element in the convex hull as min |u|_2 st. u = \sum c_i vecs[i] a

Propagates a batch of features through the model. :param features: features batch :return: model's predictions

Propagates a batch of features through the model. :param features: features batch :return: model's predictions

Determines a set of roadblock candidate where ego is located :param ego_state: class containing ego state :param map_api: map object