GR00T¶
NVIDIA's inference service over ZMQ. The service runs in NVIDIA's container; this is the client.
from strands_robots import Robot, make_policy
robot = Robot("so101", mode="sim")
policy = make_policy("zmq://gpu-box:5555", groot_version="n1.7")
robot.preflight(policy) # refuses INVALID_ACTION until observe() carries the cameras the model needs
| fact | value |
|---|---|
| provider | groot |
| wire | ZMQ request / reply, msgpack; groot_version selects the wire shape |
| frame | declared per embodiment in data_configs.json |
| in-process loaders | gone; lerobot serves GR00T N1.7 locally through lerobot |