l5kit.environment.kinematic_model module

class l5kit.environment.kinematic_model.KinematicModel

Bases: abc.ABC

Base class interface for kinematic model.

model_prefix: str

The prefix that will identify the model class

abstract reset(init_state: numpy.ndarray) → None

Reset the model state when new episode starts.

Parameters

init_state – the initial state of the ego

abstract update(input_action: l5kit.simulation.unroll.TrajectoryStateIndices) → Dict[str, numpy.ndarray]

Update the model state based on the action at a particular time-step during the episode.

Parameters

input_action – the prediction for next time-step

Returns

reward at a particular frame index (time-step) during the episode.

class l5kit.environment.kinematic_model.UnicycleModel(model_prefix: str = 'Unicycle', min_acc: float = - 0.6, max_acc: float = 0.6, min_steer: float = - 0.07853981633974483, max_steer: float = 0.07853981633974483)

Bases: l5kit.environment.kinematic_model.KinematicModel

This class is responsible for controlling kinematics using the Unicycle model.

Parameters

  • model_prefix – the prefix that will identify this model class

  • min_acc – the threshold for minimum acceleration

  • max_acc – the threshold for maximum acceleration

  • min_steer – the threshold for minimum steering

  • max_steer – the threshold for maximum steering

model_prefix: str

The prefix that will identify the model class

reset(init_state: numpy.ndarray) → None

Reset the model state when new episode starts.

Parameters

init_state – the initial state of the ego

update(input_action: numpy.ndarray) → Dict[str, numpy.ndarray]

Update the model state based on the action at a particular time-step during the episode.

Parameters

input_action – the prediction [steer, acc] for next time-step

Returns

reward at a particular frame index (time-step) during the episode.