Boston Dynamics Unveils Wheeled Atlas, Successfully Performs 360-Degree Spinning Jump

Scooter Wheels Attached to Existing Ankle Drive Units
Showcases 360-Degree Spinning Jump and Sideways Movement

Boston Dynamics, the robotics subsidiary of Hyundai Motor Group, has unveiled Atlas, a humanoid robot equipped with wheels instead of feet. The company showcased advanced maneuvers, including a 360-degree spinning jump and sideways movement, and introduced its research into testing the performance limits of its motor and whole-body control technologies.

On November 9 (local time), Boston Dynamics released a video showing Atlas’s wheel-equipped model performing a range of movements. The model has scooter wheels attached to its existing ankle drive units. In the video, Atlas quickly moves sideways after changing direction and lands after completing a full rotation in midair.

Atlas performing a 360-degree spinning jump on wheels. Boston Dynamics YouTube video.

The video also shows Atlas lifting one leg and balancing on the wheel of the other, as well as getting up by itself from a lying position on the floor and moving forward. These demonstrations showed its ability to maintain balance and control whole-body movement even as its contact conditions with the floor changed.

This wheeled version is a research platform, not a model intended for commercialization. Boston Dynamics conducted the research to determine whether its motor and whole-body control technologies operate reliably under challenging conditions. The company explained that wheeled movement is particularly suited to testing motor performance because the motors must support the robot’s actual load while delivering high torque and speed as it starts, turns, and stops.

The research also applied reinforcement learning and simulation technology. The sim-to-real approach involves repeatedly training the robot to perform various movements in a virtual environment before applying them to the actual robot. During this process, the team focused on accurately incorporating the characteristics of real motors into the virtual environment to reduce the gap between simulation and real hardware.

Atlas, a wheel-type robot, drives while changing direction. Boston Dynamics/YouTube

Boston Dynamics said that after this training process, Atlas successfully performed a 360-degree spinning jump on its first attempt.

In the video, Atlas control engineer Wu Andy said, "It was especially impressive to see the robot perform a side dash, moving sideways as it jumped during training." He added, "I was surprised and happy when the learning pipeline enabled the robot to figure out how to roll its wheels and move on its own."

Boston Dynamics said the research confirmed the importance of combining precise motor control with reinforcement learning to advance whole-body control technology for humanoid robots.

This content was produced with the assistance of AI translation services.