l5kit.environment.feature_extractor module

class l5kit.environment.feature_extractor.CustomFeatureExtractor(observation_space: gym.spaces.dict.Dict, features_dim: int = 256, model_arch: str = 'simple_gn')

Bases: stable_baselines3.common.torch_layers.BaseFeaturesExtractor

Custom feature extractor from raster images for the RL Policy.

  • observation_space – the input observation space

  • features_dim – the number of features to extract from the input

  • model_arch – the model architecture used to extract the features

forward(observations: gym.spaces.dict.Dict) torch.Tensor

Defines the computation performed at every call.

Should be overridden by all subclasses.


Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool