mirror of
https://github.com/huggingface/lerobot.git
synced 2026-08-08 17:39:44 +00:00
CarolinePascal
c112ba6957
Documents lerobot.policies to its Wave 3 narrow scope: PreTrainedPolicy, PreTrainedConfig's factory (factory.py), policies/utils.py, and for each of the 19 policy families, the full Config dataclass plus the public forward/select_action/predict_action_chunk/get_optim_params/reset surface of the main <Family>Policy class and the make_<family>_pre_post_processors factory. Per-policy internals (backbone/model building blocks, nested ProcessorStep helpers) stay out of scope and D-ignored. Fixes several real bugs found along the way: PreTrainedPolicy.forward had a literal `_summary_`/`_description_` placeholder docstring; DiffusionPolicy and VQBeTPolicy's __init__ docstrings documented a nonexistent `dataset_stats` param; XVLAPolicy.from_pretrained's docstring described a prefix-stripping behavior the code doesn't implement; XVLAAddDomainIdProcessorStep's docstring claimed the wrong default; a handful of dead `"""Input validation..."""` statements sat after the first statement in `__post_init__` (never actually docstrings) and are removed. Adds docs/source/api/policies.mdx sections for every family's Config/Policy pair, ratchets interrogate's fail-under from 55 to 58 (measured 59% with this PR), and adds the new leaf modules to check_docstrings.py's MODULES_TO_CHECK. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
178 lines
4.3 KiB
Plaintext
178 lines
4.3 KiB
Plaintext
# Policies
|
|
|
|
Every policy inherits [`PreTrainedPolicy`], which combines a `torch.nn.Module` with the Hub mixin, so any
|
|
policy can be pushed to and loaded from the Hugging Face Hub with the same two calls.
|
|
|
|
Each policy has its own guide with training recipes and results — [ACT](../act), [SmolVLA](../smolvla),
|
|
[π₀](../pi0), [π₀.₅](../pi05) and the rest are listed under Policies. To add one, see
|
|
[Adding a Policy](../bring_your_own_policies).
|
|
|
|
## PreTrainedPolicy
|
|
|
|
The abstract base class every policy subclasses. `forward` computes the training loss, `select_action`
|
|
returns one action at a time for control loops, and `predict_action_chunk` returns a full action chunk.
|
|
|
|
[[autodoc]] lerobot.policies.pretrained.PreTrainedPolicy
|
|
- forward
|
|
- predict_action_chunk
|
|
- select_action
|
|
- get_optim_params
|
|
- reset
|
|
- from_pretrained
|
|
- supports_rtc
|
|
- push_model_to_hub
|
|
- wrap_with_peft
|
|
|
|
## PreTrainedConfig
|
|
|
|
[[autodoc]] lerobot.configs.PreTrainedConfig
|
|
|
|
## make_policy
|
|
|
|
[[autodoc]] lerobot.policies.factory.make_policy
|
|
|
|
## get_policy_class
|
|
|
|
[[autodoc]] lerobot.policies.factory.get_policy_class
|
|
|
|
## make_policy_config
|
|
|
|
[[autodoc]] lerobot.policies.factory.make_policy_config
|
|
|
|
## make_pre_post_processors
|
|
|
|
[[autodoc]] lerobot.policies.factory.make_pre_post_processors
|
|
|
|
## ACT
|
|
|
|
[[autodoc]] lerobot.policies.act.modeling_act.ACTPolicy
|
|
- all
|
|
|
|
[[autodoc]] lerobot.policies.act.configuration_act.ACTConfig
|
|
|
|
## SmolVLA
|
|
|
|
[[autodoc]] lerobot.policies.smolvla.modeling_smolvla.SmolVLAPolicy
|
|
- all
|
|
|
|
[[autodoc]] lerobot.policies.smolvla.configuration_smolvla.SmolVLAConfig
|
|
|
|
## π₀ (PI0)
|
|
|
|
[[autodoc]] lerobot.policies.pi0.modeling_pi0.PI0Policy
|
|
- all
|
|
|
|
[[autodoc]] lerobot.policies.pi0.configuration_pi0.PI0Config
|
|
|
|
## π₀-FAST (PI0Fast)
|
|
|
|
[[autodoc]] lerobot.policies.pi0_fast.modeling_pi0_fast.PI0FastPolicy
|
|
- all
|
|
|
|
[[autodoc]] lerobot.policies.pi0_fast.configuration_pi0_fast.PI0FastConfig
|
|
|
|
## π₀.₅ (PI05)
|
|
|
|
[[autodoc]] lerobot.policies.pi05.modeling_pi05.PI05Policy
|
|
- all
|
|
|
|
[[autodoc]] lerobot.policies.pi05.configuration_pi05.PI05Config
|
|
|
|
## MolmoAct2
|
|
|
|
[[autodoc]] lerobot.policies.molmoact2.modeling_molmoact2.MolmoAct2Policy
|
|
- all
|
|
|
|
[[autodoc]] lerobot.policies.molmoact2.configuration_molmoact2.MolmoAct2Config
|
|
|
|
## VLA-JEPA
|
|
|
|
[[autodoc]] lerobot.policies.vla_jepa.modeling_vla_jepa.VLAJEPAPolicy
|
|
- all
|
|
|
|
[[autodoc]] lerobot.policies.vla_jepa.configuration_vla_jepa.VLAJEPAConfig
|
|
|
|
## EO-1
|
|
|
|
[[autodoc]] lerobot.policies.eo1.modeling_eo1.EO1Policy
|
|
- all
|
|
|
|
[[autodoc]] lerobot.policies.eo1.configuration_eo1.EO1Config
|
|
|
|
## LingBot-VA
|
|
|
|
[[autodoc]] lerobot.policies.lingbot_va.modeling_lingbot_va.LingBotVAPolicy
|
|
- all
|
|
|
|
[[autodoc]] lerobot.policies.lingbot_va.configuration_lingbot_va.LingBotVAConfig
|
|
|
|
## FastWAM
|
|
|
|
[[autodoc]] lerobot.policies.fastwam.modeling_fastwam.FastWAMPolicy
|
|
- all
|
|
|
|
[[autodoc]] lerobot.policies.fastwam.configuration_fastwam.FastWAMConfig
|
|
|
|
## EVO1
|
|
|
|
[[autodoc]] lerobot.policies.evo1.modeling_evo1.Evo1Policy
|
|
- all
|
|
|
|
[[autodoc]] lerobot.policies.evo1.configuration_evo1.Evo1Config
|
|
|
|
## NVIDIA GR00T
|
|
|
|
[[autodoc]] lerobot.policies.groot.modeling_groot.GrootPolicy
|
|
- all
|
|
|
|
[[autodoc]] lerobot.policies.groot.configuration_groot.GrootConfig
|
|
|
|
## X-VLA
|
|
|
|
[[autodoc]] lerobot.policies.xvla.modeling_xvla.XVLAPolicy
|
|
- all
|
|
|
|
[[autodoc]] lerobot.policies.xvla.configuration_xvla.XVLAConfig
|
|
|
|
## Multitask DiT Policy
|
|
|
|
[[autodoc]] lerobot.policies.multi_task_dit.modeling_multi_task_dit.MultiTaskDiTPolicy
|
|
- all
|
|
|
|
[[autodoc]] lerobot.policies.multi_task_dit.configuration_multi_task_dit.MultiTaskDiTConfig
|
|
|
|
## WALL-OSS
|
|
|
|
[[autodoc]] lerobot.policies.wall_x.modeling_wall_x.WallXPolicy
|
|
- all
|
|
|
|
[[autodoc]] lerobot.policies.wall_x.configuration_wall_x.WallXConfig
|
|
|
|
## Diffusion Policy
|
|
|
|
[[autodoc]] lerobot.policies.diffusion.modeling_diffusion.DiffusionPolicy
|
|
- all
|
|
|
|
[[autodoc]] lerobot.policies.diffusion.configuration_diffusion.DiffusionConfig
|
|
|
|
## Gaussian Actor
|
|
|
|
[[autodoc]] lerobot.policies.gaussian_actor.modeling_gaussian_actor.GaussianActorPolicy
|
|
- all
|
|
|
|
[[autodoc]] lerobot.policies.gaussian_actor.configuration_gaussian_actor.GaussianActorConfig
|
|
|
|
## TD-MPC
|
|
|
|
[[autodoc]] lerobot.policies.tdmpc.modeling_tdmpc.TDMPCPolicy
|
|
- all
|
|
|
|
[[autodoc]] lerobot.policies.tdmpc.configuration_tdmpc.TDMPCConfig
|
|
|
|
## VQ-BeT
|
|
|
|
[[autodoc]] lerobot.policies.vqbet.modeling_vqbet.VQBeTPolicy
|
|
- all
|
|
|
|
[[autodoc]] lerobot.policies.vqbet.configuration_vqbet.VQBeTConfig
|