The tech wizards at Boston Dynamics have been hard at work, according to an astonishing new video released by the Massachussetts-based company on Wednesday.In this video, Atlas is demonstrating policies developed using reinforcement learning with references from human motion capture and animation, reads the somewhat dry description accompanying the footage.Instead of demonstrating policies, a more accurate account would surely be, busting a slew of jaw-dropping moves thatll have you sitting straight up in your seat hollering, Woah, did that robot really just do that?!? Atlas has been impressing us for years. Who can forget the first time we saw it performing a backflip, or doing parkour all very impressive stuff. But since last year, when Atlas was relaunched as a fully electric humanoid robot with AI and machine-learning tools, the team of engineers at Boston Dynamics has taken its capabilities up a notch, to the point where its moving just like a human, in a totally natural way.But thats not all.The footage also shows Atlas doing a spot of breakdancing, which, lets be honest, if most of us tried would likely end with a slipped disc, a scream of pain, and a trip to the hospital. But Atlas performs like a pro.But its the finale that really impresses. Not even bothering with a run-up, Atlas does a perfect cartwheel. Its the kind of gleeful move that you can imagine them making after subjugating the masses.This latest work was done as part of a research partnership between Boston Dynamics and the Robotics and the Massachussetts-based AI Institute (RAI Institute), the company said.Boston Dynamics has described the latest iteration of Atlas as one of the most advanced humanoid robots ever built, adding that its now stronger, more dexterous, and more agile, and able to move in ways that exceed human capabilities.The goal is to make Atlas fit for manufacturing scenarios, helping to perform mundane tasks in a more efficient manner, freeing up human workers for more meaningful jobs. And were kind of hoping thats all it does.Editors Recommendations