simulo.Policy

The policy contract: a callable that maps observations to actions.

A policy is used both during evaluation/playback (a simulo.Player drives it) and for deployment. Calls are vectorized — a batch of observations in, a batch of actions out.

Structural mirror of simulo.core.policy.Policy.

input_dim

Policy.input_dim: Optional[int]

property

Expected observation width, or None when the policy does not declare one.

output_dim

Policy.output_dim: Optional[int]

property

Produced action width, or None when the policy does not declare one.

__call__()

Policy.__call__(observation: TensorLike) -> TensorLike

Map an observation batch (num_envs, input_dim) to actions (num_envs, output_dim).

---

Note

simulo.Policy is a mode-aware name: at submit time it resolves to a lean, torch-free stand-in so your app imports instantly; on the GPU worker it resolves to the real runtime class. The contract below is the definitive list of members every simulo.Policy provides, and the contract your own subclass fulfils.

Declared as:

@runtime_checkable
class PolicyProtocol(Protocol)