At first glance it looks like just another factory full of silently working robotic arms, but what's happening inside that building in Massachusetts may be one of the most ambitious attempts ever made to teach machines to learn like humans.

The startup Tutor Intelligence created something it calls DF1, a kind of kindergarten for robots, it is 100 bimanual manipulators working at the same time, observing, making mistakes, repeating movements and being designed by remote human operators distributed between Mexico and the Philippines.

The idea behind this is simple but gigantic, if the bear robots want to live in the real world, they need to learn in the real world, according to Josh Grustin, founder of the company, the big problem of the modern industry is that there is no internet of physical experiences for machines, while text artificial intelligences learned by reading billions of pages, robos still do not have a Wikipedia of human movement, so Tutor decided to build it manually and perhaps that is precisely where this project becomes interesting.

Instead of relying solely on digital simulations, the company put its robberies to deal with real objects, real packaging and real situations. At first, the Sony robots were clumsy, even when handling sponges or bags of chips, but after a few weeks of supervised training they began to evolve quickly.

The company's logic is almost biological. 100 robos learning simultaneously creates a kind of accelerated collective intelligence, a rare error that would take 8 hours to appear in a single robot can be identified in just 5 minutes within DF1 and that completely changes the speed of robotic evolution.

Tutor Intelligence received 34,0000 in investments and is already collaborating with companies such as Nvidia, Amazon Web Services and M Robotics, the ultimate goal is not to create just a specialized robot, but a general-purpose robotic intelligence capable of continuously learning while working.