Expanding its AI ambitions, Facebook just unveiled AI Habitat -- a simulation platform designed to train “embodied agents” like virtual robots in photo-realistic 3D environments.
For Facebook, the platform represents a significant endorsement of “embodied AI,” which uses interactive environments to ground systems’ training in the real world, rather than relying on static data sets.
In the not-too-distant future, the approach could help AI-assisted robots respond to an owner’s request to “grab my phone from the desk upstairs,” or help a visually impaired person navigate an unfamiliar subway system.
AI Habitat has an open and modular design that Facebook says is flexible enough to bring reproducibility and standardized benchmarks to this budding field.
To demonstrate AI Habitat’s abilities, Facebook’s researchers are also sharing Replica -- a data set of realistic 3D reconstructions of a staged apartment, retail store and other indoor spaces.
While AI Habitat can already run Replica’s reconstructions, the platform also works with existing 3D assets created for embodied research, such as the Gibson and Matterport3D data sets, according to Google.
To put AI Habitat through the paces, Facebook recently hosted an autonomous navigation challenge that ran on the platform, and plans to award Google Cloud credits to the winning teams at the Habitat Embodied Agents workshop later this year.
Embodied AI is not entirely new terrain for Facebook. Among other related efforts, the tech titan says it has created agents that communicate with each other to find their way through simulated NYC streets, as well as to navigate various virtual indoor environments.
With AI Habitat, Facebook is also offering Habitat-API -- a library for defining tasks such as visual navigation and answering questions.Looking into the future, Facebook sees itself pairing photo-realistic virtual environments with equally realistic avatars to create what it calls a “true social presence.”