A group from the University of London, Google DeepMind and Intrinsic introduced an artificial intelligence system called RoboBallet, coordinating the movements of some robotic hands for a few seconds. In traditional factory workshops, this requires hundreds of hours of manual programming of experts so that robots do not meet and obstacles. The results are published in robotic science.
RoboBallet turns industrial robots into a dance dance, in which each hand moves correctly and takes into account its neighboring countries. This not only stops clashes, but also creates harmony on the size of the factory
The system combines strengthening training and graphic nerve network. Put simply, whoever try different motion options, get points for points for faster and successful tasks and gradually learn how to find optimal solutions. The graph structure allows us to consider the tasks and obstacles as a network button, helping to simplify the calculation of complex coordination between robots and help avoid collisions even in densely populated seminars.
In the tests, RoboBallet deals with up to 40 tasks at the same time on eight robots, even in the arrangements that the system has never seen before. The system creates new plans hundreds of times faster than real time, allowing plants to immediately adapt to the change in the layout, new tasks or incidents of one of the robots.
Alex Lee, Associate Professor UCL, Note:
Now the coordination of some robots in the factory is how to solve a three -dimensional puzzle, in which each movement should be calculated perfectly. RoboBallet can create plans for new arrangements at a speed that cannot be achieved for one person.
The use of RoboBallet is beyond welding. In the future, the system can control robots to assemble, paint, transport parts and other tasks that require a strict series of operation. So far, RoboBallet does not take into account the robot with different capabilities or obstacles, but the AI architecture allows you to add these functions in future versions.
The project is funded by Google Deepmind and Intrinsic, and the code is opened to researchers and developers, which can accelerate the implementation of AI plan in the industry.
The main breakthrough of RoboBallet is the ability to expand: the system of understanding general coordination rules and does not remember specific cases. This allows you to control a large number of hands and adapt to any change in the factory.
With the help of RoboBallet, industrial robots can work with accuracy and consistency, this seems impossible, Matthew Matthew concluded
The first activity in the world related to two robots took place in Sydney
Xiaomi launched a self -study smartphone factory
Register and read “science” in telegram
