Implementation:Facebookresearch Habitat lab Gym Definitions

Knowledge Sources	Facebookresearch_Habitat_lab
Domains	Embodied_AI, Reinforcement_Learning
Last Updated	2026-02-15 00:00 GMT

Overview

This module registers Habitat environments with OpenAI Gym, providing factory functions to create gym-compatible Habitat environments from configuration files and pre-registering standard benchmark tasks.

Description

The module handles the integration between Habitat-Lab and OpenAI Gym's environment registration system. It operates at two levels:

Factory Functions:

• make_gym_from_config(config, dataset) -- Creates a gym environment from a Habitat config by looking up the environment class from config.env_task in the Habitat registry.
• _make_habitat_gym_env(cfg_file_path, override_options, use_render_mode) -- Internal factory that loads a config from a YAML file, optionally adds a third-person RGB sensor for rendering, and creates the gym environment.
• _get_env_name(cfg) -- Extracts the environment task name from config, handling both nested and flat config formats.
• _add_sim_sensor_to_config(config, sensor) -- Adds a simulator sensor (e.g., ThirdRGBSensorConfig) to the default agent in the config using the read_write context manager.
• _try_register(id_name, entry_point, kwargs) -- Safely registers a gym environment, skipping if already registered.

Pre-registered Tasks: The module pre-registers the following gym environments at import time via the PRE_REGISTERED_GYM_TASKS dictionary:

• Rearrange skills: CloseFridge, Pick, NavToObj, ReachState, CloseCab, OpenCab, Place, OpenFridge
• Multi-task: SetTable, Rearrange, PrepareGroceries, RearrangeEasy, TidyHouse

Each task is registered twice: once as Habitat{Name}-v0 and once as HabitatRender{Name}-v0 (with a third-person RGB sensor added). Additionally, two generic environments are registered:

• Habitat-v0 -- Generic environment supporting arbitrary configs
• HabitatRender-v0 -- Same with rendering support

Usage

Import this module to auto-register Habitat environments with Gym. Then use standard gym.make("HabitatPick-v0") calls or use make_gym_from_config for custom configurations.

Code Reference

Source Location

• Repository: Facebookresearch_Habitat_lab
• File: habitat-lab/habitat/gym/gym_definitions.py
• Lines: 1-138

Signature

def make_gym_from_config(config: "DictConfig", dataset=None) -> gym.Env:

def _make_habitat_gym_env(
    cfg_file_path: str,
    override_options: List[Any] = None,
    use_render_mode: bool = False,
) -> gym.Env:

def _try_register(id_name, entry_point, kwargs):

Import

from habitat.gym.gym_definitions import make_gym_from_config
# Or simply import the module to trigger auto-registration:
import habitat.gym.gym_definitions

I/O Contract

Inputs (make_gym_from_config)

Name	Type	Required	Description
config	DictConfig	Yes	Habitat-lab or habitat-baselines configuration (may contain "habitat" key)
dataset	object	No (default=None)	Optional dataset to pass to the environment

Inputs (_make_habitat_gym_env)

Name	Type	Required	Description
cfg_file_path	str	Yes	Path to the YAML configuration file
override_options	List[Any]	No (default=None)	List of config overrides
use_render_mode	bool	No (default=False)	Whether to add a third-person RGB sensor for rendering

Outputs

Name	Type	Description
make_gym_from_config()	gym.Env	A gym-compatible Habitat environment
_make_habitat_gym_env()	gym.Env	A gym-compatible Habitat environment from a config file

Pre-Registered Gym Environments

Gym ID	Config File
HabitatPick-v0	benchmark/rearrange/skills/pick.yaml
HabitatPlace-v0	benchmark/rearrange/skills/place.yaml
HabitatNavToObj-v0	benchmark/rearrange/skills/nav_to_obj.yaml
HabitatReachState-v0	benchmark/rearrange/skills/reach_state.yaml
HabitatOpenFridge-v0	benchmark/rearrange/skills/open_fridge.yaml
HabitatCloseFridge-v0	benchmark/rearrange/skills/close_fridge.yaml
HabitatOpenCab-v0	benchmark/rearrange/skills/open_cab.yaml
HabitatCloseCab-v0	benchmark/rearrange/skills/close_cab.yaml
HabitatRearrange-v0	benchmark/rearrange/multi_task/rearrange.yaml
HabitatRearrangeEasy-v0	benchmark/rearrange/multi_task/rearrange_easy.yaml
HabitatSetTable-v0	benchmark/rearrange/multi_task/set_table.yaml
HabitatTidyHouse-v0	benchmark/rearrange/multi_task/tidy_house.yaml
HabitatPrepareGroceries-v0	benchmark/rearrange/multi_task/prepare_groceries.yaml

Usage Examples

Basic Usage

import gym
import habitat.gym  # Triggers auto-registration

# Create a pre-registered Habitat gym environment
env = gym.make("HabitatPick-v0")
obs = env.reset()
done = False
while not done:
    action = env.action_space.sample()
    obs, reward, done, info = env.step(action)
env.close()

# Create from a custom config
from habitat.gym.gym_definitions import make_gym_from_config
from habitat import get_config

config = get_config("benchmark/rearrange/skills/pick.yaml")
env = make_gym_from_config(config)

Related Pages

• Facebookresearch_Habitat_lab_RLTaskEnv
• Facebookresearch_Habitat_lab_Config_Read_Write