Bashny.Net
How to teach a robot to cook?
Bashny.Net
Researchers at the University of Maryland and the Australian Research ICT hub announced a method that will allow robots to learn on their own cooking. In this case, the robots will use a fairly common method today - will watch videos on Youtube.
The researchers used a newly developed methods deep learning and neural networks - programs that simulate the human brain with the help of artificial neurons. Scientists have "fed" system tutorials, published on Youtube, and the system in turn identified the objects used by cooks, and tried to understand what actions are needed for cooking.
So scientists have developed a system that in the future will allow the robot to perform the most basic of kitchen tasks, and to recognize their right while watching movies.
Although it would seem, some robots already know how to cook. For example, Cooki from start Serenity performs actions in sequence, boils water adds in her pasta, roast - in short, doing everything to replace your kitchen.
Edition Motherboard took интервью a member of team of developers working on the preparation for teaching the robot movies, Cornelia Fermyuller.
How do you make a robot learn to cook? B>
There are many studies on the topic of learning robot movements, it is a well-studied area. But the work is aimed at copying movements - the robot sees the action and repeats it. In real life is more complicated, because the need to adapt to different situations. Bank or blade can be of different sizes, can be in different places at the kitchen table, the table itself may be something swamped - all this complicates the work.
The robot must be able to perceive the world. For example, "cook" need to open a jar of peanut butter, which is always in the same place. And it moved the people. It is necessary to prepare the machine for such situations.
Is the technique of deep learning and the use of clips on Youtube sufficient to overcome these problems? B>
The robot looks normal video rather than scenes shot in the laboratory. There are many ways to capture, rollers differ angle scenes, lighting, and many other characteristics. The review may obscure hand, the product can block the other objects. Here we use the technique of deep training . This is software that allows you to recognize objects, including when they are in the hand of man.
Our robots are not ready yet. But we have something that they are able to, we can pour the water and prevent something, but robots that can operate autonomously and cook, learn this on their own, have not yet done.
What are the technical obstacles you face? B>
First of all you need to solve all the above problems. We must teach the robot to perform actions in different environments, places and situations. On the other hand - while cooking anything can happen. Everything must be taken into account.
The robot must have the intelligence, the ability to reason, to identify unforeseen events and respond to them. That is to be autonomous. We continue to research work on artificial intelligence, but have not yet created it. Progress is, but it takes more time.
Returning to learning - you guessed to be able to use videos to teach a robot? B>
Users manipulate various ways. It is not enough to watch the video and get the robot to perform the same action. I and you can open a jar of peanut butter quite differently.
We solved the problem this way: you need to think not about the movement and the result. Robot action breaks into small pieces.
To open a jar of peanut butter, you must perform the following steps: move your hand towards the banks to take up the cover, turn the cover, remove the arm with a lid. Understanding these actions, this "grammar" as understanding the language - just as you can be decomposed into parts of the proposal.
When robots will watch and prepare as we? B>
Hard to say. It will happen, especially if the situation will be calm, do not run around children. But need more time. Have to deal with all kinds of perceiving things. For example, when the robot goes to the refrigerator, it should be understood that lettuce may not be home - will have to go to the store.
Source: geektimes.ru/post/244169/
The researchers used a newly developed methods deep learning and neural networks - programs that simulate the human brain with the help of artificial neurons. Scientists have "fed" system tutorials, published on Youtube, and the system in turn identified the objects used by cooks, and tried to understand what actions are needed for cooking.
So scientists have developed a system that in the future will allow the robot to perform the most basic of kitchen tasks, and to recognize their right while watching movies.
Although it would seem, some robots already know how to cook. For example, Cooki from start Serenity performs actions in sequence, boils water adds in her pasta, roast - in short, doing everything to replace your kitchen.
Edition Motherboard took интервью a member of team of developers working on the preparation for teaching the robot movies, Cornelia Fermyuller.
How do you make a robot learn to cook? B>
There are many studies on the topic of learning robot movements, it is a well-studied area. But the work is aimed at copying movements - the robot sees the action and repeats it. In real life is more complicated, because the need to adapt to different situations. Bank or blade can be of different sizes, can be in different places at the kitchen table, the table itself may be something swamped - all this complicates the work.
The robot must be able to perceive the world. For example, "cook" need to open a jar of peanut butter, which is always in the same place. And it moved the people. It is necessary to prepare the machine for such situations.
Is the technique of deep learning and the use of clips on Youtube sufficient to overcome these problems? B>
The robot looks normal video rather than scenes shot in the laboratory. There are many ways to capture, rollers differ angle scenes, lighting, and many other characteristics. The review may obscure hand, the product can block the other objects. Here we use the technique of deep training . This is software that allows you to recognize objects, including when they are in the hand of man.
Our robots are not ready yet. But we have something that they are able to, we can pour the water and prevent something, but robots that can operate autonomously and cook, learn this on their own, have not yet done.
What are the technical obstacles you face? B>
First of all you need to solve all the above problems. We must teach the robot to perform actions in different environments, places and situations. On the other hand - while cooking anything can happen. Everything must be taken into account.
The robot must have the intelligence, the ability to reason, to identify unforeseen events and respond to them. That is to be autonomous. We continue to research work on artificial intelligence, but have not yet created it. Progress is, but it takes more time.
Returning to learning - you guessed to be able to use videos to teach a robot? B>
Users manipulate various ways. It is not enough to watch the video and get the robot to perform the same action. I and you can open a jar of peanut butter quite differently.
We solved the problem this way: you need to think not about the movement and the result. Robot action breaks into small pieces.
To open a jar of peanut butter, you must perform the following steps: move your hand towards the banks to take up the cover, turn the cover, remove the arm with a lid. Understanding these actions, this "grammar" as understanding the language - just as you can be decomposed into parts of the proposal.
When robots will watch and prepare as we? B>
Hard to say. It will happen, especially if the situation will be calm, do not run around children. But need more time. Have to deal with all kinds of perceiving things. For example, when the robot goes to the refrigerator, it should be understood that lettuce may not be home - will have to go to the store.
Source: geektimes.ru/post/244169/
Tags
See also
How to teach children Thanks
As technological innovations will change everyday life
The "deep learning" will lead us to Superintelligent robots
Robot-humanoid Robotics company Hanson responds to facial expressions of the person
How to apply robots in education
10 humanoid robots created in the likeness of human abilities and emotions
Whether the robots empathy?
How to make a robot with their hands
