gerlachje's picture
Upload folder using huggingface_hub
406662d verified

Isaac Lab: Environment Suite

Using the core framework developed as part of Isaac Lab, we provide various learning environments for robotics research. These environments follow the gym.Env API from OpenAI Gym version 0.21.0. The environments are registered using the Gym registry.

Each environment's name is composed of Isaac-<Task>-<Robot>-v<X>, where <Task> indicates the skill to learn in the environment, <Robot> indicates the embodiment of the acting agent, and <X> represents the version of the environment (which can be used to suggest different observation or action spaces).

The environments are configured using either Python classes (wrapped using configclass decorator) or through YAML files. The template structure of the environment is always put at the same level as the environment file itself. However, its various instances are included in directories within the environment directory itself. This looks like as follows:

isaaclab_tasks/locomotion/
├── __init__.py
└── velocity
    ├── config
    │   └── anymal_c
    │       ├── agent  # <- this is where we store the learning agent configurations
    │       ├── __init__.py  # <- this is where we register the environment and configurations to gym registry
    │       ├── flat_env_cfg.py
    │       └── rough_env_cfg.py
    ├── __init__.py
    └── velocity_env_cfg.py  # <- this is the base task configuration

The environments are then registered in the isaaclab_tasks/locomotion/velocity/config/anymal_c/__init__.py:

gym.register(
    id="Isaac-Velocity-Rough-Anymal-C-v0",
    entry_point="isaaclab.envs:ManagerBasedRLEnv",
    disable_env_checker=True,
    kwargs={
        "env_cfg_entry_point": f"{__name__}.rough_env_cfg:AnymalCRoughEnvCfg",
        "rl_games_cfg_entry_point": f"{agents.__name__}:rl_games_rough_ppo_cfg.yaml",
        "rsl_rl_cfg_entry_point": f"{agents.__name__}.rsl_rl_ppo_cfg:AnymalCRoughPPORunnerCfg",
        "skrl_cfg_entry_point": f"{agents.__name__}:skrl_rough_ppo_cfg.yaml",
    },
)

gym.register(
    id="Isaac-Velocity-Flat-Anymal-C-v0",
    entry_point="isaaclab.envs:ManagerBasedRLEnv",
    disable_env_checker=True,
    kwargs={
        "env_cfg_entry_point": f"{__name__}.flat_env_cfg:AnymalCFlatEnvCfg",
        "rsl_rl_cfg_entry_point": f"{agents.__name__}.rsl_rl_ppo_cfg:AnymalCFlatPPORunnerCfg",
        "rl_games_cfg_entry_point": f"{agents.__name__}:rl_games_flat_ppo_cfg.yaml",
        "skrl_cfg_entry_point": f"{agents.__name__}:skrl_flat_ppo_cfg.yaml",
    },
)

Note: As a practice, we specify all the environments in a single file to avoid name conflicts between different tasks or environments. However, this practice is debatable and we are open to suggestions to deal with a large scaling in the number of tasks or environments.