Habitat
MetaHabitat is Meta's platform for embodied-AI research: Habitat-Sim, a 3D simulator that renders RGB-D and semantic camera views of scanned indoor scenes at thousands of frames per second with Bullet rigid-body physics, and Habitat-Lab, the library that defines navigation, rearrangement, instruction-following and question-answering tasks and trains agents on them. It is the navigation and home-environment simulator of the category.
Both READMEs now open with "Beyond v0.3.4 this project is no longer receiving official active development or maintenance by Meta internal teams", so the last releases (Habitat-Sim 0.3.3, Habitat-Lab 0.3.4) are the current ones. Habitat-Sim itself installs from the aihabitat conda channel, which reports only an all-time count; the PyPI habitat-sim name holds a single 2023 development upload and is not declared. The scanned scene datasets it is usually run on (HM3D, Matterport3D, Gibson) are separate downloads under their owners' terms.
Openness
5 medium confidence- license
- MIT(Habitat-Sim and Habitat-Lab code)
- source
- public
- core features withheld
- no — no paid edition
- task-data-terms
- CC-BY-NC-SA-3.0-US(the Matterport3D- and Gibson-based task datasets and trained models Meta distributes
The simulator and the task library are both MIT, with no paid edition. The navigation task datasets and trained models Meta publishes for the Matterport3D and Gibson scenes are CC BY-NC-SA, and the scanned HM3D scenes are for non-commercial research only, so commercial work uses Habitat's CC BY 4.0 ReplicaCAD scenes or brings its own.
- https://aihabitat.org/datasets/hm3d/ recorded 2026-09-27
HM3D page: "HM3D is free and available here for academic, non-commercial research."
- https://aihabitat.org/datasets/replica_cad/ recorded 2026-09-27
ReplicaCAD page: "intended for use in the Habitat simulator for embodied in-home interaction tasks such as object re-arrangement. All materials are licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0) Public License."
- https://ungh.cc/repos/facebookresearch/habitat-lab/files/main/README.md recorded 2026-09-27
Habitat-Lab README: "Habitat-Lab is MIT licensed"; "The trained models and the task datasets are considered data derived from the correspondent scene datasets"; "Matterport3D based task datasets and trained models are distributed with Matterport3D Terms of Use and under CC BY-NC-SA 3.0 US license"; "Gibson based task datasets, the code for generating such datasets, and trained models are distributed with Gibson Terms of Use and under CC BY-NC-SA 3.0 US license".
- https://ungh.cc/repos/facebookresearch/habitat-sim/files/main/LICENSE recorded 2026-09-27
LICENSE: MIT License, "Copyright (c) Meta Platforms, Inc. and its affiliates", with no added clauses; habitat-lab's LICENSE is byte-identical.
- https://ungh.cc/repos/facebookresearch/habitat-sim/files/main/README.md recorded 2026-09-27
README: "Habitat-Sim is MIT licensed"; "Beyond v0.3.4 this project is no longer receiving official active development or maintenance by Meta internal teams"; conda install from the aihabitat channel; no pricing or paid tier.
- https://ungh.cc/repos/facebookresearch/habitat-sim/releases/latest recorded 2026-09-27
Latest Habitat-Sim release v0.3.3.
Adoption
1 low confidencePyPI installs of habitat-lab, the task and training library, which links this project. The simulator installs from Meta's conda channel, which reports only an all-time total since 2020, not a monthly figure.
- https://api.anaconda.org/package/aihabitat/habitat-sim recorded 2026-09-27
Anaconda package aihabitat/habitat-sim: license MIT, latest_version 0.3.3, ndownloads 246482 (all-time), created 2020-01-23.
- https://pypi.org/pypi/habitat-lab/json recorded 2026-09-27
PyPI metadata for habitat-lab: author Meta AI Research, project_urls "GitHub repo": https://github.com/facebookresearch/habitat-lab/, license MIT License.
- https://pypistats.org/api/packages/habitat-lab/recent recorded 2026-09-27
pypistats: last_month 3,273 downloads for habitat-lab.
Capability
3 medium confidenceHabitat renders camera views of real scanned homes fast enough to train navigation agents on billions of frames, which is its strength. Its physics is Bullet on the CPU, one environment per process, so it has neither the GPU-parallel simulation nor the task library for common robot arms that set Isaac Lab well above it.
- https://ungh.cc/repos/facebookresearch/habitat-lab/files/main/README.md recorded 2026-09-27
Habitat-Lab README: tasks for navigation, rearrangement, instruction following, question answering and human following; training via imitation or reinforcement learning.
- https://ungh.cc/repos/facebookresearch/habitat-sim/files/main/README.md recorded 2026-09-27
README: "Configurable sensors (RGB-D cameras, egomotion sensing)"; "Rigid-body mechanics (via Bullet)"; "achieves several thousand frames per second (FPS) running single-threaded and reaches over 10,000 FPS multi-process on a single GPU"; "simulates a Fetch robot interacting in ReplicaCAD scenes at over 8,000 steps per second (SPS)".
Verified 2026-09-27