In an effort to build a growing database for the AI of the Facebook giant, the company recently built a series of houses to help AI learn faster.
Artificial intelligence technology now helps people a lot in life. By loading a large amount of data to learn AI, technology giants have developed smart AI machines in a remarkable way.
But not only does it take data from real life, Facebook wants more. The company has built a series of houses not used for residential purposes, but only serves the purpose of helping AI acquire knowledge.
Habitat, Facebook's new project.The company built houses not used for living, but let AI have space to practice after theoretical lessons.
Called Habitat , the houses are fully built like luxurious apartments. AI is "taught" for hundreds of hours continuously to learn how to manipulate, move in the house, avoid furniture, open doors, pull desk drawers, .
The purpose of this project is to help AI manipulate in real time. When they are put to use in the future, they will judge the situation and handle the situation in the most accurate way instead of just loading their memory of theoretical data.
By placing artificial intelligence in real houses, they will sense three-dimensional space. This is much more efficient than just loading two-dimensional images and "squeezing" them to reason.
Inside the houses built by Facebook to learn lessons AI has full furniture like real houses.
Not only familiarizing with the deep space, AI also learned how to map for any space, because the houses are organized like the real thing, including kitchen, bathroom, doors, living room, sofa,.
After spending hours "learning from reality" , the AI was able to draw a map of the space they were standing in, they could even speculate on the space obscured by furniture like how people think.
Further, Facebook's artificial intelligence technology not only draws 3D common shapes, but they can label objects through observation. For example, the AI does not identify it as a chair, but it must be a gray bench with blue pillows.
The 3D map was created when the AI observed a room in the Habitat project.
How to identify this item helps a lot for AI when they are put into practical use. By knowing the exact nature of the object, the AI makes a decision every time it needs to perform something at the highest level of accuracy.
Applied in practice, smart houses for people with disabilities will need AI with knowledge from this level upwards. AI technology will serve exactly the needs that homeowners need, smart home owners need not describe too much and still get what they want.
However, Habitat still has many drawbacks and needs more time to practice before commercialization. The AI loaded into the 'body' is a robot whose legs are wheels that make it difficult to reach every corner of the house. Robots that use depth-imaging cameras and color analyzers via RGB color systems have not yet reached the highest accuracy that the research team wants.
Different views are seen by the AI from the actual room.By looking at different wavelengths of light, they are easier to identify specific items in the room.
Moreover, the 3D map built into the computer is not completely accurate compared to the real world, meaning that the AI needs to be more physically interactive instead of just making virtual maps and judgment.
In addition to Facebook's Habitat, THOR is a similar project focusing on more physical interaction. THOR in addition to helping computers map space, the team also allows the machine to regularly interact with real objects so that it can get used to it.
Anyway, Facebook and technology giants are trying to build more advanced AI technology to support people in modern life in the future.