Spawning Multiple Assets

Spawning Multiple Assets#

Typical spawning configurations (introduced in the Spawning prims into the scene tutorial) copy the same asset (or USD primitive) across the different resolved prim paths from the expressions. For instance, if the user specifies to spawn the asset at “/World/Table_.*/Object”, the same asset is created at the paths “/World/Table_0/Object”, “/World/Table_1/Object” and so on.

However, we also support multi-asset spawning with two mechanisms:

Rigid object collections. This allows the user to spawn multiple rigid objects in each environment and access/modify them with a unified API, improving performance.
Spawning different assets under the same prim path. This allows the user to create diverse simulations, where each environment has a different asset.

This guide describes how to use these two mechanisms.

The sample script multi_asset.py is used as a reference, located in the IsaacLab/scripts/demos directory.

Code for multi_asset.py

# Copyright (c) 2022-2026, The Isaac Lab Project Developers (https://github.com/isaac-sim/IsaacLab/blob/main/CONTRIBUTORS.md).
# All rights reserved.
#
# SPDX-License-Identifier: BSD-3-Clause

"""This script demonstrates how to spawn multiple objects in multiple environments.

.. code-block:: bash

    # Usage
    ./isaaclab.sh -p scripts/demos/multi_asset.py --num_envs 2048

"""

from __future__ import annotations

"""Launch Isaac Sim Simulator first."""


import argparse

from isaaclab.app import AppLauncher

# add argparse arguments
parser = argparse.ArgumentParser(description="Demo on spawning different objects in multiple environments.")
parser.add_argument("--num_envs", type=int, default=512, help="Number of environments to spawn.")
# append AppLauncher cli args
AppLauncher.add_app_launcher_args(parser)
# demos should open Kit visualizer by default
parser.set_defaults(visualizer=["kit"])
# 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 random

from pxr import Gf, Sdf

import isaaclab.sim as sim_utils
from isaaclab.assets import (
    Articulation,
    ArticulationCfg,
    AssetBaseCfg,
    RigidObject,
    RigidObjectCfg,
    RigidObjectCollection,
    RigidObjectCollectionCfg,
)
from isaaclab.scene import InteractiveScene, InteractiveSceneCfg
from isaaclab.sim import SimulationContext
from isaaclab.sim.utils.stage import get_current_stage
from isaaclab.utils import Timer
from isaaclab.utils.assets import ISAACLAB_NUCLEUS_DIR
from isaaclab.utils.configclass import configclass

##
# Pre-defined Configuration
##

from isaaclab_assets.robots.anymal import ANYDRIVE_3_LSTM_ACTUATOR_CFG  # isort: skip


##
# Randomization events.
##


def randomize_shape_color(prim_path_expr: str):
    """Randomize the color of the geometry."""
    # get stage handle
    stage = get_current_stage()
    # resolve prim paths for spawning and cloning
    prim_paths = sim_utils.find_matching_prim_paths(prim_path_expr)
    # manually clone prims if the source prim path is a regex expression
    with Sdf.ChangeBlock():
        for prim_path in prim_paths:
            # spawn single instance
            prim_spec = Sdf.CreatePrimInLayer(stage.GetRootLayer(), prim_path)

            # DO YOUR OWN OTHER KIND OF RANDOMIZATION HERE!
            # Note: Just need to acquire the right attribute about the property you want to set
            # Here is an example on setting color randomly
            color_spec = prim_spec.GetAttributeAtPath(prim_path + "/geometry/material/Shader.inputs:diffuseColor")
            color_spec.default = Gf.Vec3f(random.random(), random.random(), random.random())


##
# Scene Configuration
##


@configclass
class MultiObjectSceneCfg(InteractiveSceneCfg):
    """Configuration for a multi-object scene."""

    # ground plane
    ground = AssetBaseCfg(prim_path="/World/defaultGroundPlane", spawn=sim_utils.GroundPlaneCfg())

    # lights
    dome_light = AssetBaseCfg(
        prim_path="/World/Light", spawn=sim_utils.DomeLightCfg(intensity=3000.0, color=(0.75, 0.75, 0.75))
    )

    # rigid object
    object: RigidObjectCfg = RigidObjectCfg(
        prim_path="/World/envs/env_.*/Object",
        spawn=sim_utils.MultiAssetSpawnerCfg(
            assets_cfg=[
                sim_utils.ConeCfg(
                    radius=0.3,
                    height=0.6,
                    visual_material=sim_utils.PreviewSurfaceCfg(diffuse_color=(0.0, 1.0, 0.0), metallic=0.2),
                ),
                sim_utils.CuboidCfg(
                    size=(0.3, 0.3, 0.3),
                    visual_material=sim_utils.PreviewSurfaceCfg(diffuse_color=(1.0, 0.0, 0.0), metallic=0.2),
                ),
                sim_utils.SphereCfg(
                    radius=0.3,
                    visual_material=sim_utils.PreviewSurfaceCfg(diffuse_color=(0.0, 0.0, 1.0), metallic=0.2),
                ),
            ],
            random_choice=True,
            rigid_props=sim_utils.RigidBodyPropertiesCfg(
                solver_position_iteration_count=4, solver_velocity_iteration_count=0
            ),
            mass_props=sim_utils.MassPropertiesCfg(mass=1.0),
            collision_props=sim_utils.CollisionPropertiesCfg(),
        ),
        init_state=RigidObjectCfg.InitialStateCfg(pos=(0.0, 0.0, 2.0)),
    )

    # object collection
    object_collection: RigidObjectCollectionCfg = RigidObjectCollectionCfg(
        rigid_objects={
            "object_A": RigidObjectCfg(
                prim_path="/World/envs/env_.*/Object_A",
                spawn=sim_utils.SphereCfg(
                    radius=0.1,
                    visual_material=sim_utils.PreviewSurfaceCfg(diffuse_color=(1.0, 0.0, 0.0), metallic=0.2),
                    rigid_props=sim_utils.RigidBodyPropertiesCfg(
                        solver_position_iteration_count=4, solver_velocity_iteration_count=0
                    ),
                    mass_props=sim_utils.MassPropertiesCfg(mass=1.0),
                    collision_props=sim_utils.CollisionPropertiesCfg(),
                ),
                init_state=RigidObjectCfg.InitialStateCfg(pos=(0.0, -0.5, 2.0)),
            ),
            "object_B": RigidObjectCfg(
                prim_path="/World/envs/env_.*/Object_B",
                spawn=sim_utils.CuboidCfg(
                    size=(0.1, 0.1, 0.1),
                    visual_material=sim_utils.PreviewSurfaceCfg(diffuse_color=(1.0, 0.0, 0.0), metallic=0.2),
                    rigid_props=sim_utils.RigidBodyPropertiesCfg(
                        solver_position_iteration_count=4, solver_velocity_iteration_count=0
                    ),
                    mass_props=sim_utils.MassPropertiesCfg(mass=1.0),
                    collision_props=sim_utils.CollisionPropertiesCfg(),
                ),
                init_state=RigidObjectCfg.InitialStateCfg(pos=(0.0, 0.5, 2.0)),
            ),
            "object_C": RigidObjectCfg(
                prim_path="/World/envs/env_.*/Object_C",
                spawn=sim_utils.ConeCfg(
                    radius=0.1,
                    height=0.3,
                    visual_material=sim_utils.PreviewSurfaceCfg(diffuse_color=(1.0, 0.0, 0.0), metallic=0.2),
                    rigid_props=sim_utils.RigidBodyPropertiesCfg(
                        solver_position_iteration_count=4, solver_velocity_iteration_count=0
                    ),
                    mass_props=sim_utils.MassPropertiesCfg(mass=1.0),
                    collision_props=sim_utils.CollisionPropertiesCfg(),
                ),
                init_state=RigidObjectCfg.InitialStateCfg(pos=(0.5, 0.0, 2.0)),
            ),
        }
    )

    # articulation
    robot: ArticulationCfg = ArticulationCfg(
        prim_path="/World/envs/env_.*/Robot",
        spawn=sim_utils.MultiUsdFileCfg(
            usd_path=[
                f"{ISAACLAB_NUCLEUS_DIR}/Robots/ANYbotics/ANYmal-C/anymal_c.usd",
                f"{ISAACLAB_NUCLEUS_DIR}/Robots/ANYbotics/ANYmal-D/anymal_d.usd",
            ],
            random_choice=True,
            rigid_props=sim_utils.RigidBodyPropertiesCfg(
                disable_gravity=False,
                retain_accelerations=False,
                linear_damping=0.0,
                angular_damping=0.0,
                max_linear_velocity=1000.0,
                max_angular_velocity=1000.0,
                max_depenetration_velocity=1.0,
            ),
            articulation_props=sim_utils.ArticulationRootPropertiesCfg(
                enabled_self_collisions=True, solver_position_iteration_count=4, solver_velocity_iteration_count=0
            ),
            activate_contact_sensors=True,
        ),
        init_state=ArticulationCfg.InitialStateCfg(
            pos=(0.0, 0.0, 0.6),
            joint_pos={
                ".*HAA": 0.0,  # all HAA
                ".*F_HFE": 0.4,  # both front HFE
                ".*H_HFE": -0.4,  # both hind HFE
                ".*F_KFE": -0.8,  # both front KFE
                ".*H_KFE": 0.8,  # both hind KFE
            },
        ),
        actuators={"legs": ANYDRIVE_3_LSTM_ACTUATOR_CFG},
    )


##
# Simulation Loop
##


def run_simulator(sim: SimulationContext, scene: InteractiveScene):
    """Runs the simulation loop."""
    # Extract scene entities
    # note: we only do this here for readability.
    rigid_object: RigidObject = scene["object"]
    rigid_object_collection: RigidObjectCollection = scene["object_collection"]
    robot: Articulation = scene["robot"]
    # Define simulation stepping
    sim_dt = sim.get_physics_dt()
    count = 0
    # Simulation loop
    while simulation_app.is_running():
        # Reset
        if count % 250 == 0:
            # reset counter
            count = 0
            # reset the scene entities
            # object
            root_pose = rigid_object.data.default_root_pose.torch.clone()
            root_pose[:, :3] += scene.env_origins
            rigid_object.write_root_pose_to_sim_index(root_pose=root_pose)
            root_vel = rigid_object.data.default_root_vel.torch.clone()
            rigid_object.write_root_velocity_to_sim_index(root_velocity=root_vel)
            # object collection
            default_pose_w = rigid_object_collection.data.default_body_pose.torch.clone()
            default_pose_w[..., :3] += scene.env_origins.unsqueeze(1)
            rigid_object_collection.write_body_pose_to_sim_index(body_poses=default_pose_w)
            default_vel_w = rigid_object_collection.data.default_body_vel.torch.clone()
            rigid_object_collection.write_body_com_velocity_to_sim_index(body_velocities=default_vel_w)
            # robot
            # -- root state
            root_pose = robot.data.default_root_pose.torch.clone()
            root_pose[:, :3] += scene.env_origins
            robot.write_root_pose_to_sim_index(root_pose=root_pose)
            root_vel = robot.data.default_root_vel.torch.clone()
            robot.write_root_velocity_to_sim_index(root_velocity=root_vel)
            # -- joint state
            joint_pos, joint_vel = (
                robot.data.default_joint_pos.torch.clone(),
                robot.data.default_joint_vel.torch.clone(),
            )
            robot.write_joint_position_to_sim_index(position=joint_pos)
            robot.write_joint_velocity_to_sim_index(velocity=joint_vel)
            # clear internal buffers
            scene.reset()
            print("[INFO]: Resetting scene state...")

        # Apply action to robot
        robot.set_joint_position_target_index(target=robot.data.default_joint_pos.torch)
        # Write data to sim
        scene.write_data_to_sim()
        # Perform step
        sim.step()
        # Increment counter
        count += 1
        # Update buffers
        scene.update(sim_dt)


def main():
    """Main function."""
    # Load kit helper
    sim_cfg = sim_utils.SimulationCfg(dt=0.005, device=args_cli.device)
    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_cfg = MultiObjectSceneCfg(num_envs=args_cli.num_envs, env_spacing=2.0, replicate_physics=True)
    with Timer("[INFO] Time to create scene: "):
        scene = InteractiveScene(scene_cfg)

    with Timer("[INFO] Time to randomize scene: "):
        # DO YOUR OWN OTHER KIND OF RANDOMIZATION HERE!
        # Note: Just need to acquire the right attribute about the property you want to set
        # Here is an example on setting color randomly
        randomize_shape_color(scene_cfg.object.prim_path)

    # Play the simulator
    sim.reset()
    # Now we are ready!
    print("[INFO]: Setup complete...")
    # Run the simulator
    run_simulator(sim, scene)


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

This script creates multiple environments, where each environment has:

a rigid object collection containing a cone, a cube, and a sphere
a rigid object that is either a cone, a cube, or a sphere, chosen at random
an articulation that is either the ANYmal-C or ANYmal-D robot, chosen at random

Rigid Object Collections#

Multiple rigid objects can be spawned in each environment and accessed/modified with a unified (env_ids, obj_ids) API. While the user could also create multiple rigid objects by spawning them individually, the API is more user-friendly and more efficient since it uses a single physics view under the hood to handle all the objects.

    init_state=RigidObjectCfg.InitialStateCfg(pos=(0.0, 0.0, 2.0)),
)

# object collection
object_collection: RigidObjectCollectionCfg = RigidObjectCollectionCfg(
    rigid_objects={
        "object_A": RigidObjectCfg(
            prim_path="/World/envs/env_.*/Object_A",
            spawn=sim_utils.SphereCfg(
                radius=0.1,
                visual_material=sim_utils.PreviewSurfaceCfg(diffuse_color=(1.0, 0.0, 0.0), metallic=0.2),
                rigid_props=sim_utils.RigidBodyPropertiesCfg(
                    solver_position_iteration_count=4, solver_velocity_iteration_count=0
                ),
                mass_props=sim_utils.MassPropertiesCfg(mass=1.0),
                collision_props=sim_utils.CollisionPropertiesCfg(),
            ),
            init_state=RigidObjectCfg.InitialStateCfg(pos=(0.0, -0.5, 2.0)),
        ),
        "object_B": RigidObjectCfg(
            prim_path="/World/envs/env_.*/Object_B",
            spawn=sim_utils.CuboidCfg(
                size=(0.1, 0.1, 0.1),
                visual_material=sim_utils.PreviewSurfaceCfg(diffuse_color=(1.0, 0.0, 0.0), metallic=0.2),
                rigid_props=sim_utils.RigidBodyPropertiesCfg(
                    solver_position_iteration_count=4, solver_velocity_iteration_count=0
                ),
                mass_props=sim_utils.MassPropertiesCfg(mass=1.0),
                collision_props=sim_utils.CollisionPropertiesCfg(),
            ),
            init_state=RigidObjectCfg.InitialStateCfg(pos=(0.0, 0.5, 2.0)),
        ),
        "object_C": RigidObjectCfg(
            prim_path="/World/envs/env_.*/Object_C",
            spawn=sim_utils.ConeCfg(
                radius=0.1,
                height=0.3,
                visual_material=sim_utils.PreviewSurfaceCfg(diffuse_color=(1.0, 0.0, 0.0), metallic=0.2),
                rigid_props=sim_utils.RigidBodyPropertiesCfg(
                    solver_position_iteration_count=4, solver_velocity_iteration_count=0
                ),
                mass_props=sim_utils.MassPropertiesCfg(mass=1.0),
                collision_props=sim_utils.CollisionPropertiesCfg(),
            ),
            init_state=RigidObjectCfg.InitialStateCfg(pos=(0.5, 0.0, 2.0)),

The configuration RigidObjectCollectionCfg is used to create the collection. It’s attribute rigid_objects is a dictionary containing RigidObjectCfg objects. The keys serve as unique identifiers for each rigid object in the collection.

Spawning different assets under the same prim path#

It is possible to spawn different assets and USDs under the same prim path in each environment using the spawners MultiAssetSpawnerCfg and MultiUsdFileCfg:

We set the spawn configuration in RigidObjectCfg to be MultiAssetSpawnerCfg:

)

# rigid object
object: RigidObjectCfg = RigidObjectCfg(
    prim_path="/World/envs/env_.*/Object",
    spawn=sim_utils.MultiAssetSpawnerCfg(
        assets_cfg=[
            sim_utils.ConeCfg(
                radius=0.3,
                height=0.6,
                visual_material=sim_utils.PreviewSurfaceCfg(diffuse_color=(0.0, 1.0, 0.0), metallic=0.2),
            ),
            sim_utils.CuboidCfg(
                size=(0.3, 0.3, 0.3),
                visual_material=sim_utils.PreviewSurfaceCfg(diffuse_color=(1.0, 0.0, 0.0), metallic=0.2),
            ),
            sim_utils.SphereCfg(
                radius=0.3,
                visual_material=sim_utils.PreviewSurfaceCfg(diffuse_color=(0.0, 0.0, 1.0), metallic=0.2),
            ),
        ],
        random_choice=True,
        rigid_props=sim_utils.RigidBodyPropertiesCfg(
            solver_position_iteration_count=4, solver_velocity_iteration_count=0
        ),
        mass_props=sim_utils.MassPropertiesCfg(mass=1.0),
        collision_props=sim_utils.CollisionPropertiesCfg(),

This function allows you to define a list of different assets that can be spawned as rigid objects. When random_choice is set to True, one asset from the list is randomly selected and spawned at the specified prim path.

Similarly, we set the spawn configuration in ArticulationCfg to be MultiUsdFileCfg:

)

# articulation
robot: ArticulationCfg = ArticulationCfg(
    prim_path="/World/envs/env_.*/Robot",
    spawn=sim_utils.MultiUsdFileCfg(
        usd_path=[
            f"{ISAACLAB_NUCLEUS_DIR}/Robots/ANYbotics/ANYmal-C/anymal_c.usd",
            f"{ISAACLAB_NUCLEUS_DIR}/Robots/ANYbotics/ANYmal-D/anymal_d.usd",
        ],
        random_choice=True,
        rigid_props=sim_utils.RigidBodyPropertiesCfg(
            disable_gravity=False,
            retain_accelerations=False,
            linear_damping=0.0,
            angular_damping=0.0,
            max_linear_velocity=1000.0,
            max_angular_velocity=1000.0,
            max_depenetration_velocity=1.0,
        ),
        articulation_props=sim_utils.ArticulationRootPropertiesCfg(
            enabled_self_collisions=True, solver_position_iteration_count=4, solver_velocity_iteration_count=0
        ),
        activate_contact_sensors=True,
    ),
    init_state=ArticulationCfg.InitialStateCfg(
        pos=(0.0, 0.0, 0.6),
        joint_pos={
            ".*HAA": 0.0,  # all HAA
            ".*F_HFE": 0.4,  # both front HFE
            ".*H_HFE": -0.4,  # both hind HFE
            ".*F_KFE": -0.8,  # both front KFE
            ".*H_KFE": 0.8,  # both hind KFE
        },

Similar to before, this configuration allows the selection of different USD files representing articulated assets.

Things to Note#

Similar asset structuring#

While spawning and handling multiple assets using the same physics interface (the rigid object or articulation classes), it is essential to have the assets at all the prim locations follow a similar structure. In case of an articulation, this means that they all must have the same number of links and joints, the same number of collision bodies and the same names for them. If that is not the case, the physics parsing of the prims can get affected and fail.

The main purpose of this functionality is to enable the user to create randomized versions of the same asset, for example robots with different link lengths, or rigid objects with different collider shapes.

Disabling physics replication in interactive scene#

By default, the flag scene.InteractiveScene.replicate_physics is set to True. This flag informs the physics engine that the simulation environments are copies of one another so it just needs to parse the first environment to understand the entire simulation scene. This helps speed up the simulation scene parsing.

However, in the case of spawning different assets in different environments, this assumption does not hold anymore. Hence the flag scene.InteractiveScene.replicate_physics must be disabled. For a full guide on the template-based cloning system including strategies and collision filtering, see Cloning Environments.

count += 1
# Update buffers
scene.update(sim_dt)

The Code Execution#

To execute the script with multiple environments and randomized assets, use the following command:

./isaaclab.sh -p scripts/demos/multi_asset.py --num_envs 2048

This command runs the simulation with 2048 environments, each with randomly selected assets. To stop the simulation, you can close the window, or press Ctrl+C in the terminal.