The Japanese company Preferred Networks has developed a helper robots control system that fully automates the cleaning process. The system includes voice recognition, through which the robot understands voice commands, as well as computer vision technologies that help the device analyze objects, lift them and put them in place. Robots that can manipulate objects find application in many spheres of human life: for example, in manufacturing and warehouses, in housework. In this case, one of the main tasks of such robots is to learn how to correctly recognize objects for further action: the object may be too fragile, or too slippery, or lie so that it cannot be lifted. This is especially true for cleaning robots, because they need not only to effectively recognize objects and take them, but also to put them in place.

Japanese engineers from Preferred Networks decided to start developing such a robot. They did not make the hardware part of the robot themselves, but chose Toyota as Humanoid Robot, which is used to help people with disabilities, as a cleaning robot; instead, the developers decided to focus on software. The computer vision system is based on a convolutional neural network that has been trained on almost two million images. Thanks to this, the system is able to efficiently recognize objects, after which the robot can identify them, understand exactly how to raise a specific object and where to place it. For example, the robot knows that the pens should lie in a glass for stationery, and the things that lie on the floor should be in the laundry basket.

nanonewsnet.ru