planet(pretrained: bool = False, overrides: object = {})Construct PlaNet's latent dynamics model — no training objective.
The architecture of Hafner et al., 2019: a four-layer stride-2 convolutional encoder, a recurrent state-space model with a 200-unit deterministic path and a 30-unit stochastic state, a mirrored transposed-convolutional decoder, and a two-layer reward head.
Model Size
Parameters
pretrainedbool= FalseNo weights are published for this family; passing
True raises
rather than returning a randomly initialised model.**overridesobject= {}Optional
PlaNetConfig field overrides. action_dim is
the one you almost always need — it is set by the environment, not
by the paper.Returns
PlaNetModelThe trunk, configured with the paper defaults and any overrides.
Notes
Reference: Hafner, Lillicrap, Fischer, Villegas, Ha, Lee, and Davidson, "Learning Latent Dynamics for Planning from Pixels", ICML, 2019 (arXiv:1811.04551).
The recurrence carries both a deterministic and a stochastic path:
Examples
>>> from lucid.models.generative.planet import planet
>>> model = planet(action_dim=6).eval()
>>> model.config.stoch_size, model.config.deter_size
(30, 200)