Habitat Lab is Meta’s modular, high-level library for end-to-end development in embodied AI. It lets researchers train agents to perform tasks in realistic indoor 3D environments: navigating to goals, rearranging objects, following instructions, answering questions, and even following or collaborating with humans.
Habitat Lab is an open-source project on GitHub with more than 3,100 stars, released under the MIT license. It sits on top of Habitat-Sim, the high-performance physics simulator from Meta AI, and provides the task definitions, agent configurations, training pipelines, and benchmarking tools around it. Researchers can define single-agent and multi-agent tasks, instantiate diverse embodied agents, from commercial robot platforms to humanoids, with configurable sensors and capabilities, and train them with imitation learning, reinforcement learning, or classical SensePlanAct pipelines. Standard metrics make results comparable across papers and labs. One honest caveat: Meta has ended official active maintenance beyond version 0.3.4, so the project now continues through community forks and contributions, though the repository still saw updates in 2026. Even so, Habitat remains one of the reference platforms for embodied AI research, with a large body of published work built on it.
Habitat Lab is free and open source under the MIT license. There is no hosted version or paid tier; you run it on your own hardware, ideally with a capable GPU for simulation and training.
Habitat Lab is aimed at robotics and embodied AI researchers, reinforcement learning practitioners, and university labs that need a reproducible platform for training and evaluating agents in simulation before transferring them to real robots.