[Question] How should I create an Articulation will unactuated joints?

[BolunDai0216]

Question

I want to create an RL environment for the Kaya robot, but I noticed that the Kaya robot has 33 joint, only 3 of them should be actuated (I think the ones on the wheels should not be actuated, otherwise the wheel is way too cheap...).

I created the config class

import omni.isaac.lab.sim as sim_utils
from omni.isaac.lab.actuators import ImplicitActuatorCfg
from omni.isaac.lab.assets import ArticulationCfg
from omni.isaac.lab.utils.assets import ISAAC_NUCLEUS_DIR

KAYA_CFG = ArticulationCfg(
    spawn=sim_utils.UsdFileCfg(
        usd_path=f"{ISAAC_NUCLEUS_DIR}/Robots/Kaya/kaya.usd",
        rigid_props=sim_utils.RigidBodyPropertiesCfg(
            rigid_body_enabled=True,
            max_linear_velocity=1000.0,
            max_angular_velocity=1000.0,
            max_depenetration_velocity=100.0,
            enable_gyroscopic_forces=True,
        ),
        articulation_props=sim_utils.ArticulationRootPropertiesCfg(
            enabled_self_collisions=False,
            solver_position_iteration_count=4,
            solver_velocity_iteration_count=0,
            sleep_threshold=0.005,
            stabilization_threshold=0.001,
        ),
    ),
    init_state=ArticulationCfg.InitialStateCfg(
        pos=(0.0, 0.0, 0.09105),
        joint_pos={"axle_0_joint": 0.0, "axle_1_joint": 0.0, "axle_2_joint": 0.0},
    ),
    actuators={
        "axle_0_joint_actuator": ImplicitActuatorCfg(
            joint_names_expr=["axle_0_joint"],
            effort_limit=400.0,
            velocity_limit=100.0,
            stiffness=0.0,
            damping=0.0,
        ),
        "axle_1_joint_actuator": ImplicitActuatorCfg(
            joint_names_expr=["axle_1_joint"],
            effort_limit=400.0,
            velocity_limit=100.0,
            stiffness=0.0,
            damping=0.0,
        ),
        "axle_2_joint_actuator": ImplicitActuatorCfg(
            joint_names_expr=["axle_2_joint"],
            effort_limit=400.0,
            velocity_limit=100.0,
            stiffness=0.0,
            damping=0.0,
        ),
    },
)

and then I made some changes to the run_articulation.py to visualize the articualtion

import argparse

from omni.isaac.lab.app import AppLauncher

# add argparse arguments
parser = argparse.ArgumentParser(
    description="Tutorial on spawning and interacting with an articulation."
)
# append AppLauncher cli args
AppLauncher.add_app_launcher_args(parser)
# parse the arguments
args_cli = parser.parse_args()

# launch omniverse app
app_launcher = AppLauncher(args_cli)
simulation_app = app_launcher.app

"""Rest everything follows."""

import omni.isaac.core.utils.prims as prim_utils
import omni.isaac.lab.sim as sim_utils
import torch
from omni.isaac.lab.assets import Articulation
from omni.isaac.lab.sim import SimulationContext

##
# Pre-defined configs
##
from kaya import KAYA_CFG  # isort:skip


def design_scene() -> tuple[dict, list[list[float]]]:
    """Designs the scene."""
    # Ground-plane
    cfg = sim_utils.GroundPlaneCfg()
    cfg.func("/World/defaultGroundPlane", cfg)
    # Lights
    cfg = sim_utils.DomeLightCfg(intensity=3000.0, color=(0.75, 0.75, 0.75))
    cfg.func("/World/Light", cfg)

    # Create separate groups called "Origin1", "Origin2"
    # Each group will have a robot in it
    origins = [[0.0, 0.0, 0.0], [-1.0, 0.0, 0.0]]
    # Origin 1
    prim_utils.create_prim("/World/Origin1", "Xform", translation=origins[0])
    # Origin 2
    prim_utils.create_prim("/World/Origin2", "Xform", translation=origins[1])

    # Articulation
    kaya_cfg = KAYA_CFG.copy()
    kaya_cfg.prim_path = "/World/Origin.*/Robot"
    kaya = Articulation(cfg=kaya_cfg)

    # return the scene information
    scene_entities = {"kaya": kaya}
    return scene_entities, origins


def run_simulator(
    sim: sim_utils.SimulationContext,
    entities: dict[str, Articulation],
    origins: torch.Tensor,
):
    """Runs the simulation loop."""
    # Extract scene entities
    # note: we only do this here for readability. In general, it is better to access the entities directly from
    #   the dictionary. This dictionary is replaced by the InteractiveScene class in the next tutorial.
    robot = entities["kaya"]
    # Define simulation stepping
    sim_dt = sim.get_physics_dt()
    count = 0
    # Simulation loop
    while simulation_app.is_running():
        # Reset
        if count % 500 == 0:
            # reset counter
            count = 0
            # reset the scene entities
            # root state
            # we offset the root state by the origin since the states are written in simulation world frame
            # if this is not done, then the robots will be spawned at the (0, 0, 0) of the simulation world
            root_state = robot.data.default_root_state.clone()
            root_state[:, :3] += origins
            robot.write_root_state_to_sim(root_state)
            # set joint positions with some noise
            joint_pos, joint_vel = (
                robot.data.default_joint_pos.clone(),
                robot.data.default_joint_vel.clone(),
            )
            robot.write_joint_state_to_sim(joint_pos, joint_vel)
            # clear internal buffers
            robot.reset()
            print("[INFO]: Resetting robot state...")
        # Apply random action
        # -- generate random joint efforts
        efforts = torch.randn_like(robot.data.joint_pos) * 1.0
        # -- apply action to the robot
        robot.set_joint_effort_target(efforts)
        # -- write data to sim
        robot.write_data_to_sim()
        # Perform step
        sim.step()
        # Increment counter
        count += 1
        # Update buffers
        robot.update(sim_dt)


def main():
    """Main function."""
    # Load kit helper
    sim_cfg = sim_utils.SimulationCfg(device="cpu")
    sim = SimulationContext(sim_cfg)
    # Set main camera
    sim.set_camera_view([2.5, 0.0, 4.0], [0.0, 0.0, 2.0])
    # Design scene
    scene_entities, scene_origins = design_scene()
    scene_origins = torch.tensor(scene_origins, device=sim.device)
    # Play the simulator
    sim.reset()
    # Now we are ready!
    print("[INFO]: Setup complete...")
    # Run the simulator
    run_simulator(sim, scene_entities, scene_origins)


if __name__ == "__main__":
    # run the main function
    main()
    # close sim app
    simulation_app.close()

This works, but for this line

efforts = torch.randn_like(robot.data.joint_pos) * 1.0

it wants me to make the efforts to be of shape (2, 33) not (2, 3), which is a bit odd to me, is this the way I should set up the articulation?

Thanks in advance!

[Ozzey]

Hi! I am facing a similar problem with a mobile manipulator robot. Did you manage to find any solution?
I think it's possible to exclude joints from Articulation in the usd file but it is causing some problems for me.

[BolunDai0216]

@Ozzey I did find a solution if you are planning to do reinforcement learning on this model.

When defining the ActionsCfg, you just need to specify the actuated joint names and Isaac Lab would take care of the rest

@configclass
class ActionsCfg:
    """Action specifications for the environment."""

    joint_efforts = mdp.JointEffortActionCfg(
        asset_name="robot",
        joint_names=["actuated_joint_name_1", "actuated_joint_name_2"],
        scale=0.25,
    )

[kellyguo11]

Hello, yes the ActionsCfg will allow you to specify the joint names. If using the direct workflow without ActionsCfg, you can slice the actions buffer to assign values on for the actuated joints. An example of this can be found at https://github.com/isaac-sim/IsaacLab/blob/main/source/extensions/omni.isaac.lab_tasks/omni/isaac/lab_tasks/direct/inhand_manipulation/inhand_manipulation_env.py#L40.