simulo.RewardComponent

One term of a task’s reward.

Per-step reward contributions from every component are summed into the task reward. Structural mirror of simulo.core.rewards.RewardComponent.

initialize()

RewardComponent.initialize(robot: RobotProtocol, device: str, num_envs: int) -> None

One-time setup: cache the robot handle and size buffers for num_envs on device.

compute()

RewardComponent.compute(
    prev_actions: TensorLike,
    terminated: Optional[TensorLike],
) -> TensorLike

Return this component’s reward contribution, shape (num_envs,).

prev_actions is the last applied action batch (for action-penalty terms); terminated marks environments ending this step (for terminal bonuses/penalties) and may be None when termination info is not available.

---

Note

simulo.RewardComponent 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.RewardComponent provides, and the contract your own subclass fulfils.

Declared as:

@runtime_checkable
class RewardComponentProtocol(Protocol)