Scientists have created robots that can see the future

Robots are now able to judge what will happen in the near and far future, the fact that they can predict the future of humans is probably no longer too difficult.

Researchers at the University. Berkeley, California has successfully developed a new learning technology that allows robots to imagine their future actions. This technology promises to help robots handle some situations or control objects that they have never tried before.

In the future, this robot technology will help self-driving cars to judge road situations or create smarter robotic assistants. Currently the first prototypes focus on developing simple learning skills.

This new technology allows robots to imagine their future actions.

The team used technology called Visual Foresight . This technology helps robots predict what cameras will see if a specific motion occurs. Although current fantasies are somewhat simple and transient, they are enough to help the robot distinguish how the object moves.

Most importantly, robots can perform this capability without the need for human help or basic physical or environmental knowledge.

This is due to the visual imagination of the robot . They are formed from the exploration and monitoring of the surrounding environment. After this process, the robot can create predictive models about the world and use this model to interact with objects it has never seen.

Sergey Levine, assistant professor at Berkeley's Department of Electronic Engineering and Computer Science said: "Like the way we imagine how actions interact with objects in the environment, the method It also allows the robot to visualize how different actions will affect the world around, which helps robots to anticipate skills that respond quickly to complex real-life situations. .

A test that judges what might happen in the future of robots.(Photo of UC Berkeley).

The core element of the system is the technology of deep learning , based on predicting repeating videos or dynamic neural advection (DNA) patterns . The DNA model predicts how pixels in a picture will move from one frame to the next. Therefore, this model relies heavily on robot actions. Now DNA is improving, helping robots predict more complex actions.

Are robots capable of self-learning and collecting data superior to humans?

With the new technology, the robot can push an object on the table, then use the predictive model and choose the form of motion to move the object to a specific location. Robots use models from CCTV, then learn by themselves and teach themselves how to avoid objects or push objects around obstacles.

Frederik Ebert, a graduate student in Levine's lab, believes robots are superior to humans in self-learning skills without the need for others to teach. In particular, a robot system has the ability to collect huge amounts of data, serving for learning and training skills to interact with objects.

The great advantage of the video prediction system is that continuous observations are automatically collected. Contrary to computer vision methods, requiring people to label thousands, millions of images, creating a video prediction model only requires regular videos.

The current ability of robots is still limited but its skills are all self-learning knowledge.(Illustration).

Levine said: "Children can learn languages ​​by playing with toys, moving, grasping them and many other activities. Therefore the purpose of this study is to assist a robot to do this. Similarly: it is about how the world works through interactions.

The current ability of robots is still limited but its skills are all self-learning knowledge. Thus, the robot can predict complex physical interactions with objects that it has never touched before, just by building interactive models through observation. "

Scientists at the University. Berkeley is continuing to study how to improve and control predictions through video. At the same time, the team is considering developing more methods to help robots get more video data, especially for complex tasks such as picking and placing objects on soft or leather objects. like fabric, rope .