EEmbodied AI Hub
ProjectAdvanced

Habitat Lab

Meta embodied AI platform focused on navigation, manipulation, and scene understanding.

Official / sourceBack to Project
SimulationRLOpen source

Overview

Habitat Lab is Meta’s embodied AI platform focused on navigation, rearrangement, and scene understanding inside photorealistic 3D scans and synthetic scenes. It is the standard when your problem is “agent in a house” rather than “arm on a table”.

Expect a different stack: simulators, dataset scenes (Matterport, HM3D, etc.), and metrics like SPL. Manipulation support exists but the community gravity remains navigation and embodied perception.

Who it is for

Embodied navigation and scene-understanding researchers.

Key highlights

  • Photoreal indoor scene ecosystem
  • Standard navigation metrics and baselines
  • Scalable simulation for embodied agents

When to use

  • Indoor navigation, VLN, rearrangement in homes

When not to use

  • Contact-rich industrial arm control as primary goal

Getting started

  1. 1Install Habitat-Sim + Habitat-Lab matching versions.
  2. 2Download a small scene dataset and run pointnav.
  3. 3Only then scale to larger scans and multi-task agents.

How it compares

Caveats & pitfalls

  • Scene downloads are large; plan disk early.

Content reviewed 2026-07-23