Same Positronic inference protocol used by every model in the leaderboard. Drop the snippet below into any Python control loop.

from positronic.inference import RemotePolicy

policy = RemotePolicy(
    host="warehouse-pick-prod.ep-9k2a.serverless.nebius.com",
    token=os.environ["NEBIUS_ENDPOINT_TOKEN"],
)

while robot.is_active():
    obs = robot.get_observation()        # cameras + proprio
    actions = policy(obs, instruction="pick the red bin")
    robot.execute(actions)               # 4-step chunk @ 30 Hz

Stream a sample episode through the endpoint and visualise the predicted action chunks against ground truth.