AI Habitat (the simulator for robots)

2019: Announcement

In the middle of June, 2019 the Facebook company uploaded publicly the AI Habitat simulator (together with source codes) which is intended for training of robots that those could be guided in realistic conditions.

According to the VentureBeat edition that the home robot understood voice commands (for example, could check whether there is a notebook in other room, and could bring the device to the owner), it is necessary to integrate several forms of artificial intelligence: computer vision, a natural language understanding, training with a reinforcement, etc.

Facebook released the open AI simulator for training of robots in the realistic environment

For consolidation of such technologies the AI Habitat system is also created. It can process more than 10 thousand frames per second on one graphic processor that significantly quicker in comparison with other simulators.

As soon as the AI algorithm passed training and test in the simulator, it can be moved to the physical platforms (robots) working in the real world.

To AI Habitat follows from documentation that the program consists of a stack of three modular levels, each of which can be configured or replaced for work with different types of AI agents, training methods, protocols of assessment and Wednesdays. Separation of these layers distinguishes the platform from other simulators which development can complicate separation of parameters for reuse of components or comparison of results.

In addition to the engine simulator, the library of high-level AI algorithms of API for navigation of robots, following to their instructions and answers to questions is a part of AI Habitat.

The AI Habitat simulator was released for creation of a unified environment and problem solving of standardization of complex researches in community of robotics and artificial intelligence. For this purpose Facebook also released the PyTorch Hub system.^[1]

Robotics

Notes