Gym registration

Step 2 – Register Gym Environments

Next, you register two Gym environments so that RSL-RL (and Gym-compatible tools) can create your task by name.
Gym registration is the “glue” that connects a string ID (like G1-Stand-Flat-v0) to your Isaac Lab environment configuration and PPO settings.

You will register:

• G1-Stand-Flat-v0 – training environment
• G1-Stand-Flat-Play-v0 – play/visualization environment

Registration happens inside the g1_stand extension, not in Isaac Lab core.

---

Steps

Paste registration code into __init__.py

1. Open this file: source/g1_stand/g1_stand/tasks/manager_based/g1_stand/__init__.py
2. Replace its contents with the code in the block below (or paste the block into the file).

import gymnasium as gym

from . import agents

##
# Register Gym environments.
##

# Training environment: many parallel envs, used for RL training.
gym.register(
    id="G1-Stand-Flat-v0",
    entry_point="isaaclab.envs:ManagerBasedRLEnv",
    disable_env_checker=True,
    kwargs={
        "env_cfg_entry_point": f"{__name__}.g1_stand_env_cfg:G1StandFlatEnvCfg",
        "rsl_rl_cfg_entry_point": f"{agents.__name__}.rsl_rl_ppo_cfg:G1StandFlatPPORunnerCfg",
    },
)

# Play / visualization environment: few envs, no corruption, suitable for rollout.
gym.register(
    id="G1-Stand-Flat-Play-v0",
    entry_point="isaaclab.envs:ManagerBasedRLEnv",
    disable_env_checker=True,
    kwargs={
        "env_cfg_entry_point": f"{__name__}.g1_stand_env_cfg:G1StandFlatPlayEnvCfg",
        "rsl_rl_cfg_entry_point": f"{agents.__name__}.rsl_rl_ppo_cfg:G1StandFlatPPORunnerCfg",
    },
)

---

What this does

Here, “environment” means the RL task (robot, scene, and rewards), not your Python/conda environment.

• id
  Names the environment in Gym’s global registry:

• "G1-Stand-Flat-v0" – training

• "G1-Stand-Flat-Play-v0" – play

• entry_point
  Points to the Isaac Lab manager-based RL environment wrapper:

• "isaaclab.envs:ManagerBasedRLEnv"

• kwargs.env_cfg_entry_point
  Tells Isaac Lab which environment config class to use:

• Training env: G1StandFlatEnvCfg

• Play env: G1StandFlatPlayEnvCfg

• kwargs.rsl_rl_cfg_entry_point
  Tells Isaac Lab which RSL-RL runner config to use:

• G1StandFlatPPORunnerCfg (defined in the agents module)

Important

The strings inside env_cfg_entry_point and rsl_rl_cfg_entry_point must match the actual import paths and class names. If you rename files or classes, update these strings accordingly.

Checklist

After updating __init__.py:

• IDs are exactly G1-Stand-Flat-v0 and G1-Stand-Flat-Play-v0.
• Entry-point strings match: G1StandFlatEnvCfg / G1StandFlatPlayEnvCfg and G1StandFlatPPORunnerCfg.
• You’ll confirm it worked in Step 4 when you run train.py and training starts without “env not found” errors.