Nvidia has developed a robot that learns to perform tasks while observing a person

Usually, all industrial robots operate on the same principle. They are asked exactly what to do. And they execute this predetermined algorithm until the operator implements another command in them. As a rule, such industrial devices are not allowed to work close to those fragile creatures that programmed them. But the team of Nvidia scientists from Seattle found a solution to this problem, and made the first robot, which learns from the example of man. Inside, he has only a bundle of neural networks and a “dictionary” that allows him to describe what is happening around. You can read the scientific work here , and yesterday Nvidia published a video with examples of tasks on its YouTube channel.

All decisions for the robot are made by trained neural networks that understand their tasks based on a demonstration. They know how to observe the environment, generate a program for themselves, and then execute it. The robot takes into account the relationship between the objects (for example, it understands that if you remove the cube from below, the whole structure will collapse), creates whole plans for itself (for example, “I want to carefully remove the lower cube”), and then puts them into action (to remove the lowest cube - you must first clear all the previous ones). Neural networks are trained entirely in simulation, and in the real world, scientists only check the quality of their work by installing them in a particular robot. In the case of cubes, it was industrial robot Baxter ( The Baxter ), because his hands can perform the same function as that of the human hand.

Dieter Fox, head of robotics at Nvidia and a professor at the University of Washington, says the team wants to create the next generation of robots that can work safely in close proximity to humans. But for this, such robots must be able to identify people, distinguish them from the environment, monitor their activities and understand when they need help. Dieter expects to see such robots in small industries and in private homes, especially among people with disabilities. They are able to adapt to new situations on the go, and can work without a special operator trained to configure them.

A self-learning algorithm can be taught to pass games without problems - by simply repeating the same segment many times, adjusted for error. But Fox says that training is not suitable for the robot. He works in the real world, so a much wider space of solutions is available to him, and in case of an error, the result can be disastrous. Therefore, the task of the team was to train the neural network to exactly follow the example of a person, and if an unexpected deviation from such a program occurs, to understand that an error has occurred and try to eliminate it.

So far, the tasks posed to Baxter are as simple as possible. Differ cubes from each other in color and shape. Look at what a person does to them. Repeat all operations after him. This can be done by a three-year-old child. But for robotics, this is an important step in an attempt to create a universal robot that can learn new "tricks" without the intervention of a team of programmers.

And already at this stage, not everything is as smooth as it might seem. For example, the developers are very proud that the system was able to independently distinguish the red machine from the red cube, although no one had shown the machine to it before in the real world. Or, for example, it’s hard for the robot to feel the depth now, it has flat “eyes” -chambers, but he understands very well when the cube has not been placed on top of his friend, and he himself guesses how this can be fixed so that everything works out the same way as person.

A group of scientists (by the way, the Russian - Artyom Molchanov was also a member of it ) reached the point that just one demonstration is enough for the machine to repeat all the actions. Moreover, all the commands that the robot formulates for itself in its neural network brain are easily readable by humans. "Put yellow on red," "Move green to blue." Someone who has never encountered such a robot, if necessary, can easily come up and “straighten his brains”.

To train neural networks, the team mainly used synthetic data from a simulated environment. Given the speed of moving the robot arm, in the real world, networks would have to learn for years, not to mention the fact that you can break a car. Stan Burchfield, who led this project, says that creating free simulations close to the real world in which algorithms can learn from their mistakes is the only way robots can learn fast enough. Therefore, Nvidia is engaged in this development: the company seems that their hardware is ideally suited for such tasks. An important component of training is the visual aspect. Machines must understand how a person looks, and how those objects on which you have to work differ from each other. Nvidia’s experience with hardware and graphics software, according to Burchfield,

Now the team is engaged in the creation of more photorealistic simulations to make it easier for neural networks to transition to the real world, and is expanding the scope of tasks that they can remember.

PS When you buy Nvidia cards on Newegg and Amazon, you save up to 30%. And Pochtoy.com delivers them from the USA to Russia. When registering with us with the Geektimes code - $ 7 to your account.