Habitat

AI Habitat is an embodied AI simulation platform developed by Meta (FAIR). It provides a high-performance, physics-enabled 3D simulator for training and evaluating embodied AI agents in photorealistic indoor environments before transferring skills to the real world. Habitat consists of two main components: Habitat-Sim (the core high-performance physics-enabled 3D simulator) and Habitat-Lab (a modular high-level library for end-to-end development in embodied AI). Habitat supports tasks including navigation, rearrangement, instruction following, question answering, and human-following across single and multi-agent setups. The platform supports 3D scans of indoor and outdoor spaces (HM3D, MatterPort3D, Gibson, Replica), CAD models, and a diverse set of embodied agents including commercial robots and humanoids. It enables training via imitation learning, reinforcement learning, and SensePlanAct pipelines with human-in-the-loop interaction. Habitat 3.0 is the latest major release. The library has been a key platform for embodied AI research, enabling the Habitat Challenge competitions. Note: Beyond v0.3.4, Habitat-Lab is no longer receiving active development from Meta but remains available for community forking and development.