Adding a new embodiment or scene

The Isaac-Lab extension at compass/rl_env/exts/mobility_es/ defines the embodiments (H1, Carter, Spot, G1, Digit, …) and scenes COMPASS trains on. Adding your own is a registration exercise — the gin configs and run.py driver pick it up once the entry exists.

Adding new embodiments

To add a new robot embodiment, follow these steps:

1. Create robot configuration

• Follow the Isaac Lab instructions to add a new robot articulation configuration in compass/rl_env/exts/mobility_es/mobility_es/config/robots.py.

• Ensure proper joint configurations, collision properties, and physics parameters are set.

2. Create environment configuration

• Create a new RL environment class that inherits from GoalReachingEnvCfg in compass/rl_env/exts/mobility_es/mobility_es/config/env_cfg.py.

• Override the following components as needed:

  Component	Description
  scene.robot	Set robot articulation configuration
  scene.camera	Configure camera settings
  actions	Define action item for the embodiment
  observations	Configure RL observations
  events	Configure RL events
  rewards	Configure RL rewards
  terminations	Configure RL terminations

Note

Each embodiment should define its own action term as a low-level controller that maps velocity commands to joint positions. We have pre-defined action terms for different robot embodiments under compass/rl_env/exts/mobility_es/mobility_es/mdp/action that can be re-configured for new embodiments.

3. Register and verify the environment

• Import your new environment in run.py.

• Verify it is functional by running run.py with the new environment.

Examples of existing embodiments

• H1 humanoid robot

• Carter mobile robot

• Spot quadruped robot

• G1 humanoid robot

• Digit humanoid robot

Adding new scenes

1. Create scene configuration

• Follow the Isaac Lab scene tutorial to add a new scene configuration in compass/rl_env/exts/mobility_es/mobility_es/config/environments.py. Inherit from EnvSceneAssetCfg with proper parameters like pose_sample_range, env_spacing, replicate_physics, etc.

2. Add an occupancy map (optional)

We use an occupancy map for collision-free pose sampling. It’s optional but recommended for improving sampling efficiency, especially in tight scenes.

Generate the map directly from the scene USD with the bundled CLI:

${ISAACLAB_PATH}/isaaclab.sh -p scripts/generate_omap_from_usd.py path/to/scene.usd

By default this writes <scene>.png and occupancy_map.yaml to <usd_dir>/omap/. Use --cell-size, --z-min / --z-max, --bounds, or --out-dir to override the defaults; see --help for the full flag list.

You can register the YAML path explicitly in OMAP_PATHS in compass/rl_env/exts/mobility_es/mobility_es/config/environments.py, but it’s not required: when the scene has no entry, the collision checker falls back to looking for an omap/occupancy_map.yaml next to the scene’s USD — which is exactly where the generator drops it.

The legacy interactive flow via Isaac Sim’s occupancy-map UI still works if you prefer authoring a map by hand. Save the YAML/PNG into the same <usd_dir>/omap/ location and the loader will pick it up.

See the Auto OMap from USDs handbook page for the deeper generator + verifier reference.

3. Register and verify the scene

• Import your new scene in run.py.

• Verify the scene is functional by running run.py with the new scene by setting env_cfg.scene.environment = MyNewSceneCfg().