Robots learn to work together, like in a dance

It is now possible for robots working together in obstacle-filled spaces like factory assembly lines to learn to 'coexist' and coordinate their movements without colliding with each other or with surrounding objects, like dancers in a well-choreographed dance . This is the result of a study led by the London-based company Google DeepMind and published in the journal Science Robotics, which developed a new Artificial Intelligence algorithm called 'RoboBallet '. The algorithm takes just seconds to program the robots' activities , whereas human programmers currently require hundreds of hours of work, and will therefore make work in factories much faster and more efficient .

RoboBallet allows a robotic brain to be trained using reinforcement learning , which involves learning through trial and error and assigning rewards for correctly completed tasks. After just a few days of training, researchers led by Matthew Lai saw that the algorithm was able to program the activities of eight robotic arms in a matter of seconds, even for complex, previously unseen environments , completing up to 40 tasks— an achievement far beyond the capabilities of previous systems.

"RoboBallet transforms industrial robotics into a choreographed dance ," says Lai, "in which each arm moves with precision, purpose, and awareness of its teammates: it's not just about avoiding accidents, but about achieving harmony on a large scale. For the first time," the researcher continues, "we can automate the complex planning of multiple robots with the grace and speed of a dance, making factories more adaptable, efficient, and intelligent."