Atlas pulls off a 360-degree spinning jump on wheels instead of feet ... Hyundai advances robot control technology

[Financial News] Hyundai Motor Group's humanoid robot Atlas successfully completed a jump with a 360-degree midair rotation, fitted with wheels instead of feet. It also performed challenging maneuvers, such as supporting its body on just one wheel and rapidly changing direction. Hyundai Motor Group is stepping up efforts to secure a competitive edge in physical AI technology by using AI reinforcement learning to improve the robot's motor control performance and movement stability.
On the 9th (local time), Hyundai Motor Group robotics affiliate Boston Dynamics released a research video showing its humanoid robot Atlas, fitted with wheels instead of feet, performing a range of challenging maneuvers.
In the video, Atlas moves quickly on its wheels, changes direction, and performs a midair rotation before landing. It also lifts one leg, balances on a single wheel on the other side, then gets itself up from the floor and moves.
The wheeled Atlas unveiled this time is a research platform for testing the performance limits of its motors and whole-body control technology, not a product intended for commercialization. The focus was on attaching scooter wheels to the existing ankle actuators and checking how stably the robot moves under extreme conditions.
In particular, during wheeled movement—starting, turning and stopping—the motors must support the robot's weight while continuously delivering high torque and speed. This makes it a suitable setting for testing motor performance and control technology.
Boston Dynamics used AI reinforcement learning in this research. It had the robot repeatedly learn a range of movements in a virtual environment and applied Sim-to-real technology, which accurately incorporates the characteristics of real motors into simulations. The technology is designed to ensure that movements learned in a virtual environment can also be performed reliably by a real robot.
Using this approach, Atlas successfully performed, on its first attempt with a real robot, a 360-degree spinning jump modeled on human movement. Boston Dynamics said it had confirmed that combining precise motor control with reinforcement learning plays an important role in advancing robots' whole-body control technology.
Andy Wu, an Atlas control engineer at Boston Dynamics, said, “I was particularly impressed to see the robot move sideways while jumping during the learning process,” adding, “I was surprised and pleased when the robot figured out on its own how to move by rolling its wheels.”
[email protected] Kim Dong-ho Reporter