While Tesla will constantly promote ever more advanced versions of Optimus, the humanoid robot it’s working on—at least versions lacking human help behind the scenes—there is another robot in development you might be less familiar with: the Boston Dynamics Atlas series. What had been a rather bulky, chunky, yet agile robot, the Atlas II has slimmed down and been made to look even more humanlike and equally more maneuverable. While Hyundai has owned Boston Dynamics for four years, Toyota—via its Toyota Research Institute—has come in to provide new artificial intelligence to help make the Atlas II and future humanoid robots act more like humans do and potentially replace us squishy beings for dangerous and overly repetitive tasks.
Back in 2021, Hyundai purchased an 80 percent share in Boston Dynamics from SoftBank—which took ownership of it from Google in 2017. Hyundai was already using Boston Dynamics’ Spot, the doglike robot, for industrial inspection and predictive maintenance at its facilities, and the new goal was to start deploying Atlas—the humanoid robot already made famous for its dance moves and parkour skills—to further roles at Hyundai plants in the future. If you’ve been watching any of the development videos on Atlas II and seen the unmistakable Hyundai “H” logo on engine covers and other auto parts used, that’s why.
But it seems that Hyundai isn’t doing this alone anymore. The Toyota Research Institute works on developing software for automated driving and robotics, and it recently showed off its latest development through Atlas II. While being generally labeled as AI, the better description is Large Behavior Model, or LBM.
Rather than simply programming the bot to do a single repetitive task, the LBM in Atlas makes it behave more like a human and opens it up for more dynamic and nearly on-the-fly tasks. The program looks at the entire body of the robot and figures out how to perform tasks while keeping balance at the same time. Before, most humanoid robots would need to perform a balance adjustment before switching over to its task, which is why you’d see Atlas and other bots stop before continuing after something changed, be it a part falling out or something moving its original position.
Part of the demonstration of the LBM programming sees Atlas open a container with two flaps containing Spot parts. While Atlas begins to look at what it needs to do to accomplish the task, an engineer will take a hockey stick and flip one of the flaps closed. Without hesitation, it bends down and opens the flap to begin grabbing parts. Most of the demonstrations of Atlas to this point have also only shown what it can do with a single hand for a single task, but this latest video shows that Atlas can work with both hands, even just to grab a part that fell out of its grasp without much stoppage in getting its work done. It can even bend down and pull the container back into position without the need to wait until a balance positioning has been performed first,