RoboLab-120

RoboLab-120 is NVIDIA's 120-task manipulation suite on Isaac Lab. The tasks are pick-place and tidying problems. Each task is labeled with the skill it tests: semantics, spatial reasoning, color, counting, sorting, or stacking. Your policy receives a shoulder view, a wrist view, and a language instruction, and drives a DROID arm with absolute joint targets.

Benchmark	Tasks	Episodes per task	Total episodes
robolab-120	120	10	1200

Contract

Observation

Each step, the benchmark sends an observation with these fields. If Manifold has adapters for the difference, it converts the observation to the form that your policy signature declares. Then it passes the observation to your policy.

Field	Values	Meaning
state["ee_pose"]	x, y, z	position in meters
	qx, qy, qz, qw	rotation as a quaternion, scalar last
state["joint_pos"]	j0 to j6	joint angles in radians
	gripper	continuous in [0, 1], 0 opens
instruction		the task prompt, a string
sensors["over_shoulder_left"]		640x360 color image, uint8
sensors["wrist"]		640x360 color image, uint8

Action

Return actions in the exact format that your policy signature declares. Manifold will attempt to convert your policy's output to match what the benchmark accepts.

Compatible policy signatures

Change the values to match your policy. If the policy signature below matches what your policy natively requires, then it is guaranteed to be compatible with RoboLab-120.

ProprioceptionFormat

End effector pose

Joint positions

Joints, then gripper [j0, ..., j6, gripper]

CameraOrientationSize

widthheight

widthheight

Show policy signature

POLICY_SIGNATURE = PolicySignature(
    action_space=JointActionSpace(
        dof=7,
        gripper=GripperFormat.BINARY_OPEN_LOW,
        delta=False,
    ),
    proprioception=Proprioception(
        ee_pose=EEObservationSpec(rotation=RotationFormat.QUATERNION),
        joint_pos=JointObservationSpec(dof=7, gripper=GripperFormat.UNSIGNED_OPEN_LOW),
    ),
    cameras=(
        Camera(
            name="over_shoulder_left",
            shape=(360, 640, 3),
            mount=Mount.SCENE,
            orientation=CameraOrientation.UPRIGHT,
        ),
        Camera(
            name="wrist",
            shape=(360, 640, 3),
            mount=Mount.WRIST,
            orientation=CameraOrientation.UPRIGHT,
        ),
    ),
    instruction=True,
)

Debugging and testing

To check your policy end to end before a full run, submit it against debug-robolab-120, a small debug version of this benchmark.

For the full walkthrough, see Serve your policy.