The landscape of humanoid robotics is rapidly advancing, with key players like Boston Dynamics and Toyota Research Institute (TRI) pushing the boundaries of what these machines can achieve. While Tesla's Optimus garners significant attention, the Boston Dynamics Atlas series, particularly the refined Atlas II, represents a formidable contender in the race towards human-like automation. Initially a bulky yet agile robot, Atlas II has undergone a significant transformation, becoming slimmer, more human-like, and remarkably more maneuverable.

Hyundai's ownership of Boston Dynamics since 2021, after acquiring an 80 percent stake from SoftBank (which previously obtained it from Google in 2017), has been a pivotal factor in this evolution. Hyundai had already integrated Boston Dynamics' Spot robots for industrial inspections and predictive maintenance. The strategic goal was to extend this integration to the Atlas humanoid robot, leveraging its well-known dance moves and parkour skills for future roles within Hyundai's manufacturing plants. The visible Hyundai "H" logo on Atlas II components in recent development videos confirms this ongoing collaboration.

However, Hyundai's journey in enhancing Atlas II is now augmented by the expertise of the Toyota Research Institute. TRI, renowned for its work in automated driving and robotics software, has introduced a groundbreaking artificial intelligence solution for Atlas II, referred to as a Large Behavior Model (LBM). This LBM goes beyond traditional programming for repetitive tasks; it enables Atlas to behave with greater human-like adaptability, facilitating more dynamic and nearly spontaneous actions. The system intelligently considers the robot's entire body, allowing it to perform tasks while simultaneously maintaining balance – a stark contrast to older humanoid robots that required distinct balance adjustments before initiating or resuming tasks.

A recent demonstration vividly illustrates the LBM's capabilities. As Atlas began to open a two-flap container for Spot parts, an engineer intervened, closing one of the flaps with a hockey stick. Without hesitation, Atlas responded by bending down to reopen the flap and continued grabbing parts. Furthermore, previous demonstrations often showcased Atlas performing single-handed tasks. The latest video highlights its ability to work with both hands, even retrieving a dropped part without significant interruption. It can also reposition the container in one fluid motion, eliminating the need for a separate balance positioning step. This remarkable achievement in robotics has been developed in a remarkably short timeframe, with the partnership between Boston Dynamics and TRI commencing in October 2024.

The overarching objective of this advanced development is to deploy Atlas II effectively in existing operational environments. Traditional programming, which lacks the dynamic neural networks that consider the robot's whole body, would make scaling such robots into necessary roles incredibly challenging. With LBM, the training process is significantly streamlined, potentially requiring only a single human demonstration for a robot to learn a task. Russ Tedrake, senior vice president of LBM at TRI, noted that "As the LBMs get stronger, they require less and less demonstrations to achieve more and more robust behaviors." While a future where robots handle all menial tasks remains distant, innovations like the Large Behavior Model are rapidly accelerating its arrival. It is conceivable that within the next decade, entire automotive production lines could be staffed by Atlas II robots.