Using the Interactive Scene

So far in the tutorials, we manually spawned assets into the simulation and created object instances to interact with them. However, as the complexity of the scene increases, it becomes tedious to perform these tasks manually. In this tutorial, we will introduce the scene.InteractiveScene class, which provides a convenient interface for spawning prims and managing them in the simulation.

At a high-level, the interactive scene is a collection of scene entities. Each entity can be either a non-interactive prim (e.g. ground plane, light source), an interactive prim (e.g. articulation, rigid object), or a sensor (e.g. camera, lidar). The interactive scene provides a convenient interface for spawning these entities and managing them in the simulation.

Compared the manual approach, it provides the following benefits:

• Alleviates the user needing to spawn each asset separately as this is handled implicitly.
• Enables user-friendly cloning of scene prims for multiple environments.
• Collects all the scene entities into a single object, which makes them easier to manage.

In this tutorial, we take the cartpole example from the Interacting with an articulation tutorial and replace the design_scene function with an scene.InteractiveScene object. While it may seem like overkill to use the interactive scene for this simple example, it will become more useful in the future as more assets and sensors are added to the scene.

The Code

This tutorial corresponds to the create_scene.py script within scripts/tutorials/02_scene.

# 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 use the interactive scene interface to setup a scene with multiple prims.
~~~bash
# Usage
uv run python scripts/tutorials/02_scene/create_scene.py --num_envs 32
~~~
"""
"""Parse the command-line arguments first."""
import argparse
from typing import TYPE_CHECKING
from isaaclab.app import add_launcher_args, launch_simulation
# add argparse arguments
parser = argparse.ArgumentParser(description="Tutorial on using the interactive scene interface.")
parser.add_argument("--num_envs", type=int, default=2, help="Number of environments to spawn.")
# append simulation launcher cli args
add_launcher_args(parser)
# parse the arguments
args_cli = parser.parse_args()
"""Rest everything follows."""
import torch
import isaaclab.sim as sim_utils
from isaaclab.assets import ArticulationCfg, AssetBaseCfg
from isaaclab.scene import InteractiveSceneCfg
from isaaclab.sim import SimulationContext
from isaaclab.utils import configclass, instantiate, replace
if TYPE_CHECKING:
    from isaaclab.scene import InteractiveScene
##
# Pre-defined configs
##
from isaaclab_assets import CARTPOLE_CFG  # isort:skip
@configclass
class CartpoleSceneCfg(InteractiveSceneCfg):
    """Configuration for a cart-pole 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))
    )
    # articulation
    cartpole: ArticulationCfg = replace(CARTPOLE_CFG, prim_path="{ENV_REGEX_NS}/Robot")
def run_simulator(sim: sim_utils.SimulationContext, scene: "InteractiveScene"):
    """Runs the simulation loop."""
    # Extract scene entities
    # note: we only do this here for readability.
    robot = scene["cartpole"]
    # Define simulation stepping
    sim_dt = sim.get_physics_dt()
    count = 0
    # Simulation loop
    while sim.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_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)
            # set joint positions with some noise
            joint_pos, joint_vel = (
                robot.data.default_joint_pos.torch.clone(),
                robot.data.default_joint_vel.torch.clone(),
            )
            joint_pos += torch.rand_like(joint_pos) * 0.1
            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 robot state...")
        # Apply random action
        # -- generate random joint efforts
        efforts = torch.randn_like(robot.data.joint_pos.torch) * 5.0
        # -- apply action to the robot
        robot.set_joint_effort_target_index(target=efforts)
        # -- 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."""
    # Configure the simulation
    sim_cfg = sim_utils.SimulationCfg(device=args_cli.device)
    # Launch the simulator runtime that the configuration needs
    with launch_simulation(sim_cfg, args_cli):
        # Initialize the simulation context
        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 = CartpoleSceneCfg(num_envs=args_cli.num_envs, env_spacing=2.0)
        scene = instantiate(scene_cfg)
        # 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 function
    main()

The Code Explained

While the code is similar to the previous tutorial, there are a few key differences that we will go over in detail.

Scene configuration

The scene is composed of a collection of entities, each with their own configuration. These are specified in a configuration class that inherits from scene.InteractiveSceneCfg. The configuration class is then passed to the scene.InteractiveScene constructor to create the scene.

For the cartpole example, we specify the same scene as in the previous tutorial, but list them now in the configuration class CartpoleSceneCfg instead of manually spawning them.

@configclass
class CartpoleSceneCfg(InteractiveSceneCfg):
    """Configuration for a cart-pole 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))
    )
    # articulation
    cartpole: ArticulationCfg = replace(CARTPOLE_CFG, prim_path="{ENV_REGEX_NS}/Robot")

The variable names in the configuration class are used as keys to access the corresponding entity from the scene.InteractiveScene object. For example, the cartpole can be accessed via scene["cartpole"]. However, we will get to that later. First, let’s look at how individual scene entities are configured.

Similar to how a rigid object and articulation were configured in the previous tutorials, the configurations are specified using a configuration class. However, there is a key difference between the configurations for the ground plane and light source and the configuration for the cartpole. The ground plane and light source are non-interactive prims, while the cartpole is an interactive prim. This distinction is reflected in the configuration classes used to specify them. The configurations for the ground plane and light source are specified using an instance of the assets.AssetBaseCfg class while the cartpole is configured using an instance of the assets.ArticulationCfg. Anything that is not an interactive prim (i.e., neither an asset nor a sensor) is not handled by the scene during simulation steps.

Another key difference to note is in the specification of the prim paths for the different prims:

• Ground plane: /World/defaultGroundPlane
• Light source: /World/Light
• Cartpole: {ENV_REGEX_NS}/Robot

As we learned earlier, Omniverse creates a graph of prims in the USD stage. The prim paths are used to specify the location of the prim in the graph. The ground plane and light source are specified using absolute paths, while the cartpole is specified using a relative path. The relative path is specified using the ENV_REGEX_NS variable, which is a special variable that is replaced with the environment name during scene creation. Any entity that has the ENV_REGEX_NS variable in its prim path will be cloned for each environment. This path is replaced by the scene object with /World/envs/env_{i} where i is the environment index.

Scene instantiation

Unlike before where we called the design_scene function to create the scene, we now create an instance of the scene.InteractiveScene class from the configuration object. The configuration names the class it builds in its scene.InteractiveSceneCfg.class_type attribute, so ~isaaclab.utils.instantiate constructs the scene from it. While creating the configuration instance of CartpoleSceneCfg we specify how many environment copies we want to create using the num_envs argument. This will be used to clone the scene for each environment.

# Design scene
scene_cfg = CartpoleSceneCfg(num_envs=args_cli.num_envs, env_spacing=2.0)
scene = instantiate(scene_cfg)

The configuration class CartpoleSceneCfg can be defined at the top of the script, but the scene.InteractiveScene class works on the USD stage of the running simulator. The class_type attribute resolves the class lazily, so the scene class is only imported when it is constructed inside the with launch_simulation(...) block, once the simulator runtime has been launched.

Accessing scene elements

Similar to how entities were accessed from a dictionary in the previous tutorials, the scene elements can be accessed from the InteractiveScene object using the [] operator. The operator takes in a string key and returns the corresponding entity. The key is specified through the configuration class for each entity. For example, the cartpole is specified using the key "cartpole" in the configuration class.

# Extract scene entities
# note: we only do this here for readability.
robot = scene["cartpole"]

Running the simulation loop

The rest of the script looks similar to previous scripts that interfaced with assets.Articulation, with a few small differences in the methods called:

• assets.Articulation.reset ⟶ scene.InteractiveScene.reset
• assets.Articulation.write_data_to_sim ⟶ scene.InteractiveScene.write_data_to_sim
• assets.Articulation.update ⟶ scene.InteractiveScene.update

Under the hood, the methods of scene.InteractiveScene call the corresponding methods of the entities in the scene.

The Code Execution

Let’s run the script to simulate 32 cartpoles in the scene. We can do this by passing the --num_envs argument to the script.

uv run python scripts/tutorials/02_scene/create_scene.py --num_envs 32

This should open a stage with 32 cartpoles swinging around randomly. You can use the mouse to rotate the camera and the arrow keys to move around the scene.

In this tutorial, we saw how to use scene.InteractiveScene to create a scene with multiple assets. We also saw how to use the num_envs argument to clone the scene for multiple environments.

There are many more example usages of the scene.InteractiveSceneCfg in the tasks found under the isaaclab_tasks extension. Please check out the source code to see how they are used for more complex scenes.