Now, in order to teach a robot something new, it is often required to spend a lot of human effort and time on the learning process itself. But NVIDIA proposes to significantly change this approach. Their new deep machine learning system will allow robots to learn on their own by simply observing people.
The new system is built on the basis of interconnected neural networks. Some are responsible for perception, others for planning, others for data management and processing, and so on. Thanks to the interconnection of various systems and their integration with control systems, motion and object recognition, robots can learn by simply observing people. And also retrain to perform new tasks. As stated by NVIDIA Chief Scientist Stan Burchfield, “In a production environment, robots perfectly repeat the same actions, but they are not able to adapt to the changes occurring around them, and they do not learn, performing only a given program. And in order to teach the robot to perform a new task, you have to invite an expert who will reprogram the robot, which is not cheap."
The development of the NVIDIA company is aimed at so that even non-specialists can teach the robot new techniques.
“We are now at a stage where, with the help of new technologies, we can create, in fact, limitless amounts of pre-tagged data, and then develop and test algorithms. All this will help enable us to create systems that will adapt to any conditions.”
The corporation even released a video demonstrating the technology. A person moves objects on it, and, "looking at him", the robot learns to do the same. According to NVIDIA employees, the technology is only "at the beginning." Therefore, on the one hand, it is interesting, but on the other hand, it is a little scary, what else can robots learn from us in the future. The full video demo is available below.
Vladimir Kuznetsov