Load an Exported Policy

Load the exported package with auto-detection first.

from physicalai.inference import InferenceModel

model = InferenceModel("./exports/act_policy")

Then compute one action.

model.reset()
action = model.select_action(observation)

Load directly from the Hugging Face Hub with a repo id.

model = InferenceModel.from_pretrained("OpenVINO/act-fp16-ov")

Pin a revision (branch, tag, or commit SHA) for reproducible loads.

model = InferenceModel.from_pretrained(
    "OpenVINO/act-fp16-ov",
    revision="main",
)

If necessary, select the backend explicitly.

model = InferenceModel(
    "./exports/act_policy",
    backend="openvino",
    device="CPU",
)

Use chunk prediction when a runtime owns the queueing and timing.

chunk = model.predict_action_chunk(observation)

Do not build robot-loop timing around select_action(). Use PolicyRuntime when the policy is driving hardware.