# AI Habitat

> AI Habitat is an open-source embodied-AI simulation platform from Meta AI Research. It is software you
> install and run locally, not a hosted service: there is no HTTP API, no API key, no base URL and no
> pricing. Two packages carry the platform — habitat-sim, a high-performance C++/Python 3D simulator
> (10,000+ FPS rendering; 8,000+ robot steps/second), and habitat-lab, a modular Python library for
> defining embodied-AI tasks, training agents and benchmarking them. Both are MIT licensed.

IMPORTANT — maintenance status: both READMEs carry a first-party notice stating that "beyond v0.3.4
this project is no longer receiving official active development or maintenance by Meta internal teams"
and inviting the community to fork and continue development independently. v0.3.4 was released
2026-05-07. Treat Habitat as stable-but-unmaintained upstream software when planning an adoption.

Generated by the API Evangelist enrichment pipeline on 2026-08-30 from apis.yml and the artifacts in
this repository. AI Habitat does not publish its own /llms.txt (https://aihabitat.org/llms.txt, 404).

## Getting started

- [AI Habitat home](https://aihabitat.org/): Project site — overview, papers, team, datasets, challenges.
- [Concepts and terminology](https://aihabitat.org/docs/concepts/): The vocabulary of the platform — agent, sensor, scene, episode, task, measurement.
- [Resource index](https://aihabitat.org/docs/resource_index/): Curated index of every doc, tutorial, dataset and paper.
- [Habitat-Sim docs](https://aihabitat.org/docs/habitat-sim/): Simulator reference, including generated class docs and the C++ API.
- [Habitat-Lab docs](https://aihabitat.org/docs/habitat-lab/): Task/training library reference.
- [Browser demo](https://aihabitat.org/demo): WebAssembly build of the simulator that runs in the browser with no install. Live, but no longer produced from the current build (the JS build was removed in habitat-sim v0.3.3).

## Install

- Recommended (conda): `conda install habitat-sim withbullet -c conda-forge -c aihabitat`
- Training library: `pip install -e habitat-lab`, then `pip install -e habitat-baselines` for the RL/IL baselines.
- From source: `pip install . --no-build-isolation` — see [BUILD_FROM_SOURCE.md](https://github.com/facebookresearch/habitat-sim/blob/main/BUILD_FROM_SOURCE.md).
- Caution on PyPI: `pip install habitat-sim` resolves to a 0.2.4 dev build from 2023-04-03, three tagged releases behind the source. Use conda or a source build.

## Command line

- `python -m habitat_sim.utils.datasets_download --uids habitat_test_scenes --data-path /path/to/data/` — fetch test scenes.
- `viewer /path/to/scene.glb` — compiled interactive viewer, on PATH after `pip install .`.
- `python examples/viewer.py --dataset data/replica_cad/replicaCAD.scene_dataset_config.json --scene apt_1` — Python viewer.
- `python examples/example.py --scene ...` — non-interactive smoke test.

## Standards this platform speaks

- Gym / Gymnasium environment interface: `import habitat.gym; env = gym.make("HabitatRenderPick-v0")`.
- URDF for robot models (Fetch, Franka, AlienGo).
- glTF/GLB (and .ply) for scene and object assets.
- Bullet for rigid-body and articulated dynamics.
- Hydra/OmegaConf YAML for application configuration.

## Source and community

- [habitat-sim on GitHub](https://github.com/facebookresearch/habitat-sim): Simulator source, releases, issues.
- [habitat-lab on GitHub](https://github.com/facebookresearch/habitat-lab): Task/training library source, releases, issues.
- [PARTNR planner](https://github.com/facebookresearch/partnr-planner): Human-robot collaboration benchmark built on Habitat.
- [Discussions forum](https://github.com/facebookresearch/habitat-lab/discussions): The community support channel.
- [Releases (habitat-sim)](https://github.com/facebookresearch/habitat-sim/releases) and [Releases (habitat-lab)](https://github.com/facebookresearch/habitat-lab/releases): The dated change record; there is no CHANGELOG.md.
- [Security policy](https://github.com/facebookresearch/habitat-lab/security/policy): Report vulnerabilities through the Meta Bug Bounty program, never as a public issue or PR.

## Datasets and benchmarks

- [Datasets](https://aihabitat.org/datasets/) — HM3D, HM3D-Semantics, ReplicaCAD; also HSSD, MatterPort3D, Gibson, Replica. Each carries its own licence and access terms, held by the dataset publisher rather than by Meta.
- [ai-habitat on HuggingFace](https://huggingface.co/ai-habitat): Datasets and models.
- [Habitat Challenge](https://aihabitat.org/challenge/): Benchmark competitions (2019-2023) hosted on EvalAI.
- [Habitat 3.0](https://aihabitat.org/habitat3/): Humans, avatars and robots sharing one simulated environment.

## API Evangelist artifacts in this repository

- apis.yml — the APIs.json profile for AI Habitat.
- packages/ai-habitat-packages.yml — PyPI/conda/source distributions with versions and publish dates.
- lifecycle/ai-habitat-lifecycle.yml — versioning and the end-of-maintenance notice.
- changelog/ai-habitat-changelog.yml — structured release history back to v0.3.0.
- cli/ai-habitat-cli.yml — the documented command-line surface.
- sandbox/ai-habitat-sandbox.yml — browser demo and named test fixtures.
- conformance/ai-habitat-conformance.yml — domain standards (Gym, URDF, glTF, Bullet, Hydra).
- data-model/ai-habitat-data-model.yml — entity graph over the configuration and observation objects.
- json-schema/, json-structure/, examples/, vocabulary/, json-ld/ — schemas and examples for SimulatorConfig, AgentConfig, SensorSpec, Observation, Episode, NavigationGoal, TaskConfig.
- security/ — domain security probes and the vulnerability-disclosure policy.
- plans/ and rate-limits/ — both measured zero, with the evidence for why.
- well-known/ai-habitat-well-known.yml — /.well-known probe record (all 404).

## What is NOT here

There is no OpenAPI, AsyncAPI, GraphQL SDL, gRPC .proto, WSDL, MCP server, A2A agent card, OAuth scope
set, webhook catalog or error-code registry for AI Habitat, and none is missing by oversight — the
platform has no network protocol surface for any of them to describe. Probes are recorded in
well-known/ai-habitat-well-known.yml and conformance/ai-habitat-conformance.yml.