Exploring Environment Variations

The earlier environment examples chose one fixed HDR background for each launch. An Arena variation instead samples a registered environment property automatically. In this guide, Arena varies the HDR background, light intensity, and wrist-camera position. Before connecting a trained policy, you can explore these variations with the zero_action policy. The robot stays still while the environment loads and renders, so no model weights are required.

The gallery below previews supported lighting, camera, and background effects.

Lighting direction, color, temperature, and intensity variations.

Wrist-camera intrinsics and extrinsics variations.

The same task with different HDR backgrounds.

Run the example

The repository includes a DROID pick-and-place example that enables three variations: the HDR background, light intensity, and wrist-camera position. Its zero-action policy keeps the robot still, and five environment rebuilds make the visual changes easy to inspect.

Configuration file (droid_pnp_variations_experiment.yaml)

# Copyright (c) 2026, The Isaac Lab Arena Project Developers (https://github.com/isaac-sim/IsaacLab-Arena/blob/main/CONTRIBUTORS.md).
# All rights reserved.
#
# SPDX-License-Identifier: Apache-2.0

runs:
  variations_demo:
    environment:
      type: pick_and_place_maple_table
      enable_cameras: true
      embodiment: droid_rel_joint_pos
    policy:
      type: zero_action
    rollout_limit:
      num_steps: 10
    # Number of times the environment is rebuilt with different variation values.
    num_rebuilds: 5
    # The variations to enable for this run.
    variations:
      light.hdr_image.enabled: true
      light.intensity.enabled: true
      droid_rel_joint_pos.camera_extrinsics_wrist_camera.enabled: true

Start or enter the Base Docker container from the repository root:

./docker/run_docker.sh

Then run the example inside the container:

python isaaclab_arena/evaluation/experiment_runner.py \
  --viz kit \
  --experiment_config isaaclab_arena_environments/experiment_configs/droid_pnp_variations_experiment.yaml

The viewport will show the environment being rebuilt with different backgrounds and light intensities:

The example at 5x playback speed. The wrist-camera position also changes, but that change is not visible from the external viewport.

The background and light intensity are sampled before each environment build. The wrist-camera position is sampled when the environment resets. Arena calls these build-time and run-time variations. See Build-time and run-time variations for details.

Explore other variations

Use the same YAML file with --list_variations to see the variations available for its environment:

python isaaclab_arena/evaluation/experiment_runner.py \
  --experiment_config isaaclab_arena_environments/experiment_configs/droid_pnp_variations_experiment.yaml \
  --list_variations

See Variations for the available variations and their configuration options.

See Arena Experiments for configuration rules and local and OSMO execution.

Next steps

Continue to Running a Real Policy to replace the zero-action policy with a trained policy. To measure which variations are associated with success or failure, follow the Evaluation workflow.