Hold Still

Reference policy, task and request: for smoke tests and as examples of what a customer writes.

HoldStill

HoldStill(rate: float = 10.0, horizon: int = 4)

Starts at zero with the grippers open and targets the joints where they are: horizon rows at rate Hz, a new chunk every tick.

Attributes

rate

horizon

Methods

initialize(station: Station_Spec, task: Task_Spec) -> Station_JointPose

infer(observation: Station_Observation) -> Station_ActionChunk | None

reset() -> None

HOLD_STILL_TASK

HOLD_STILL_TASK = Task_Spec(name='hold_still', description='Hold the arms still at the start pose.', reset_instruction='Clear the workspace; nothing to set up.', max_rollout_time=5.0, instruction=None, scoring=Task_ScoreSheet(success=Task_ScoreItem('still', 'Did the arms stay still?', Task_ScoreKind.Binary)))

The task a smoke test asks for: nothing to set up, one yes/no question.

HOLD_STILL_REQUEST

HOLD_STILL_REQUEST = JobRequest(task=HOLD_STILL_TASK, policy=HoldStill(), episodes=1)

A complete request, as a customer would upload one.