projects.gym_baselines.experiments.mujoco.gym_mujoco_walker2d_ddppo#
GymMuJoCoWalkerConfig#
class GymMuJoCoWalkerConfig(GymMuJoCoPPOConfig)
GymMuJoCoWalkerConfig.create_model#
| @classmethod
| create_model(cls, **kwargs) -> nn.Module
We define our ActorCriticModel agent using a lightweight
implementation with separate MLPs for actors and critic,
MemorylessActorCritic.
Since this is a model for continuous control, note that the
superclass of our model is ActorCriticModel[GaussianDistr]
instead of ActorCriticModel[CategoricalDistr], since we'll use
a Gaussian distribution to sample actions.