se3labs

Client library for evaluating a robot policy on an SE3 Labs station. Your model runs on your machine; the station streams observations to it over one gRPC stream, executes the actions it returns under the station's safety layer, and records everything. Nothing here touches hardware: the only dependencies are grpcio and protobuf.

pip install se3labs

Run a reference policy against the job id your SE3 Labs contact gives you, and poll the job's progress:

se3labs eval run    --job job_… --address relay.example.com:7443 --policy hold
se3labs eval status --job job_… --address relay.example.com:7443

Wrap your own model:

from se3labs.eval.policy import client
from se3labs.eval.policy.proto import policy_session_pb2 as pb


class MyPolicy:
    def manifest(self) -> pb.ClientManifest:
        return client.default_manifest(model="my-model", version="1", max_horizon=8)

    def reset(self, reset: pb.Reset) -> None:
        pass

    def act(self, observe: pb.Observe) -> list[pb.Action]:
        targets = {a.name: (list(a.joint_pos_rad), None) for a in observe.observation.arms}
        return [client.joint_action(targets) for _ in range(8)]


client.serve_policy("job_…", "relay.example.com:7443", MyPolicy())

Or, with that class saved in my_policy.py, without writing the last line:

se3labs eval run --job job_… --address relay.example.com:7443 --policy my_policy:MyPolicy

Full documentation: https://se3labs.github.io/YAM-Station/

Protocol version: se3labs.eval.policy.PROTOCOL_VERSION. A release that changes it cannot talk to a station expecting the old one.