Google wants robots to acquire new skills by learning from each other

Google wants robots to acquire new skills by learning from each other
16:34 12 Oktyabr 2016
Ölkə mətbuatı
A- A+

Google has recently been working in the field of so-called “cloud robotics”. This phenomenon, when the robots trained to independently carry out some action, can share the “experience” with other robots, just by passing information by any available method of communication. This principle of learning avoids the time of re-programming, or so-called “retraining”, when the production technology of new tasks.
The essence of “cloud robotics” is the following: it is based on neural networks that define and store the sequence of executed actions, are responsible for the processes of automaticity and transfer of information. In General, for what we call experience. The robots based on neural networks you can put any task, and artificial brain will find solutions. In the future, when you perform these steps several times, the robot will develop an optimal algorithm that will be able to pass other cars, and they will use and improve it without starting each time from scratch .
Scientists from Google’s Research tested the algorithm on three kinds of robots performing different tasks: opening doors, examining objects on a tray and a modified version of the first experiment, when the robot was trained independently, and ruled by a man with the subsequent task is to improve the acquired skills.
In the first case, the machine left a lot of time to understand, then to open door to touch the handle, turn it and push down on the door. But all subsequent robots used this algorithm, passing the time of training.
In the experiment with the tray of the machine was left to fend for themselves within a few hours of studying the causal relationships between objects (for example: tea — Cup — sugar: what to do with it, it is obvious to us, the robots also had to “learn”).
Experiment number three after training, the robot operator was at the mercy of the “collective consciousness”, which quickly found the optimal solution, characterized by different initial positions of the manipulators and the end result, ustarevshim the product of manipulation.
The most interesting moment was when one of the robots, forced open the door, which was installed to handle a completely different type. The machine coped with the task.
Why it’s needed, in addition to build theories about the rise of the machines? It’s simple: like the acceleration of the learning process will enable the industrial robots to begin to perform complex tasks much faster than with a traditional approach.