dreamer_world_model(pretrained: bool = False, overrides: object = {})Construct Dreamer with all three of its objectives.
Same trunk as dreamer, wrapped with the world-model bound of
Hafner et al., 2019 and the two behaviour losses of Hafner et al.,
2020 — an actor maximising over imagined
trajectories and a critic regressing onto it.
Model Size
Parameters
pretrainedbool= FalseNo weights are published for this family; passing
True raises.**overridesobject= {}Optional
DreamerConfig field overrides. horizon,
lambda_ and discount shape the imagined return;
action_dim is set by the environment.Returns
DreamerForWorldModelingThe trunk plus the objectives.
Notes
Reference: Hafner, Lillicrap, Ba, and Norouzi, "Dream to Control: Learning Behaviors by Latent Imagination", ICLR, 2020 (arXiv:1912.01603).
The three losses take three separate optimisers over the parameter
groups the model exposes — see DreamerForWorldModeling.
Examples
>>> from lucid.models.generative.dreamer import dreamer_world_model
>>> model = dreamer_world_model(action_dim=6).eval()
>>> len(model.actor_parameters()) > 0
True