Cleaning a floor, drying a cup, taking a glass from a closet … Simple operations that may be impossible for elderly, sick or disabled people.
In the not too distant future these categories of people could be joined by robots that can help them in these and many other household chores. But how to teach robots what to do? What instructions should you give to machines that are interacting in environments that are different from one another in terms of furnishings, dimensions and location of spaces?
The problem is not at all simple and has recently been addressed by a team of researchers at the Toyota Research Institute who are working on a home robot.
This robot, a cart with a “head” with optical sensors and two arms, already knows how to perform different operations, but it is not able to generalize the instructions it receives. It is not able to abstract the concept “I take a glass from the cabinet and fill it with water” applying it to any glass or cabinet.
To resolve this issue, the Toyota team decided to entrust the robot’s training to human beings in the flesh who, through special visors similar to that for virtual reality, can “see” through the eyes of the machine. The study was published on arXiv.
The instructor, through a special menu that appears in the display, can instruct the machine to perform the most correct movement depending on the situation.
It’s easy to say “take”. The simple gesture of grabbing a cup, depending on the initial position of the cup, where it will be taken, what it contains and so on can be declined in different movements: hold it by the handle, take it by the edge, or tighten it like a glass.
Likewise, putting something in a locker requires different movement sequences depending on the height of the cabinet, the size of the item to be stored, the type of door opening that can be swing or sliding and so on.
From a practical point of view what the human instructor does is to “mark” the various objects with an electronic label and associate them with the different movements, so that it is easy for robots to know what to do in the most diverse situations.
As the modern food delivery robots, this approach allows the robot to accumulate different experiences and skills in a relatively short time, and allows him to learn to act in the most correct way according to the scenario he is facing.
The domestic robot, specified by Toyota, is not, at least for now, destined to arrive in our homes: it is an exclusively scientific project used by researchers to explore the potentials and limits of these technologies.
© Copyright BizTechNews.net